I Seen Long Ago, Given the Wrong Name
Open almost any dinosaur book today and the animals have colors. Chestnut tails. Long feathers patterned in black and white. Dark backs and pale bellies. Twenty years ago, every one of those colors came from an artist.
Fossils are gray or brown, the colors of the rock itself. Every brushstroke of color was an act of imagination, and the artists knew it. The gray-green dinosaur skin in the 1993 film Jurassic Park was no different.
The verdict then was brisk: color does not fossilize. There were two fairly solid reasons for saying so.
First, the colors you see on a fossil mostly belong to minerals, oxide films, or carbonaceous residues. They have nothing to do with how the animal looked alive. Second, even if its original pigment molecules had once been there, tens of millions of years would have broken them down long ago. The fossil's survival does not mean its original optical properties survived with it.
Color was the standard example of a particular kind of absence: something survives, but what survives does not carry the property you want to ask about. The bones are there. The feather impressions are there. The color is not. This essay is about what happened to that example. It was partly overturned, and what happened after the overturning matters more than the overturning itself.
Yet there had long been counterexamples to the claim that “color can leave no evidence in principle.” The ink sacs of cephalopods, the group that includes squid and octopuses, were well known for surviving as black organic residues in fossils. As early as 1973, researchers compared the infrared spectra of fossil ink and modern melanin.
By 2012, more direct chemical methods had yielded fairly strong evidence of eumelanin in two Jurassic ink sacs at least 160 million years old. Ink is an extreme case: pigment as concentrated as pigment gets, already packaged in a sac.
But it was enough to show that molecules such as melanin have unusually high preservation potential. In 2009, researchers in Britain removed the ink sac of a squid-like animal from a Jurassic fossil more than 160 million years old. They ground up the ink, mixed it with an ammonia solution, and used it to draw the very animal it had come from.
The metallic sheen of fossil beetles and moths offers an entirely different route. If their nanoscale layers survive, an electron microscope can measure the thickness of the layers and the distances between them. An optical model can then work backward to the wavelengths they once reflected.
From the structure of wing scales in 47-million-year-old moths, researchers calculated that the upper surfaces of their forewings had been yellow-green in life, with a color that did not change with the viewing angle. Studies of fossil beetles between 15 million and 47 million years old also turned up a problem we will encounter repeatedly: the color a fossil has now need not be the color it had alive. The beetles' nanolayers were actually quite well preserved. What had changed were the optical properties of the shell material itself. The reflected colors had shifted toward the red end of the spectrum: what had been blue became green; what had been yellow became orange.
So, before dinosaur color entered the discussion, what had long been missing was not “any information about color.” It was a chain of calibration: a dependable way to map the material or nanostructure in a fossil back to the color category it had belonged to in life.
And the most important part of the prehistory is this: people had already seen these little bodies under the microscope.
First, what are we looking for? Variations in melanin give human hair and skin their lighter and darker shades. Inside cells, it is packed into tiny granules called melanosomes, only a micrometer or two long, tens of times smaller than the width of a human hair.
Melanin comes in two kinds. Eumelanin produces blacks and grays; pheomelanin produces reds and yellows, including the red in red hair. In bird feathers, the granules containing eumelanin are mostly long and slender, like sausages. Those containing pheomelanin are mostly short and rounded, like meatballs.
Messel, in Germany, is an abandoned oil-shale pit. Forty-seven million years ago it was a volcanic lake, and its fossils are so well preserved that an animal's last meal can still be seen inside its stomach. In the 1980s, the site narrowly escaped being turned into a garbage dump.
In 1983, researchers described large numbers of micrometer-scale rods and spheres in Messel fossils, including feathers and hair. At the time, they were interpreted as microorganisms involved in the decay of soft tissue.
In 1995, two other researchers developed the idea further. They proposed that feathers had retained their outlines through an organic film formed by bacteria responsible for decomposition. Later, the material surrounding the little bodies was also interpreted as extracellular polymers secreted by bacteria.
In other words, the evidence had been sitting under the microscope all along. It had been seen, photographed, and published. It had simply been filed under another name. This matters if we are to understand the next twenty years: the same set of images can sustain two entirely different worlds.
The real turn came in 2008. Four researchers reexamined an Early Cretaceous feather with light and dark bands from Brazil's Crato Formation. The formation is lake-bed limestone more than 110 million years old, quarried locally for paving; fossils often turn up inside its slabs. They compared this feather with black feathers from living birds.
Here is what they saw. The dark bands were densely packed with elongated, carbon-containing ellipsoids roughly one to two micrometers long. In the light bands, little remained beyond the feather's impression. The bodies' size, shape, and alignment along the barbs and barbules looked more like melanosomes than a randomly spread bacterial film.
Two papers published in 2010 took a further step: from “one feather may have preserved melanosomes” to systematic sampling of dinosaur specimens, classification into color categories, and reconstruction of patterns across the whole body. It was a large step. The moment it was taken, the trouble began.
II Twenty-Nine Points
In January 2010, a paper appeared online in Nature. Its subjects were Sinosauropteryx, Sinornithosaurus, Confuciusornis, and an isolated feather. All came from the lakes and volcanic ash of western Liaoning more than 120 million years ago, where feathered dinosaurs and ancient birds had been buried in great numbers: the Jehol Biota. The researchers used scanning electron microscopy to examine micrometer- and submicrometer-scale bodies on the surfaces of feathers and filaments, then compared their sizes and shapes with melanosomes in modern feathers.
They divided the bodies into two types: longer rods and shorter, nearly spherical forms. In living birds, the first correspond to eumelanin, the second to pheomelanin.
Different parts of the Sinosauropteryx tail contained different types of bodies. The authors therefore interpreted the tail's visible rings as distinct pigment zones, proposing alternating reddish-brown and lighter colors. Discovered in Liaoning in 1996, Sinosauropteryx was the first non-avian dinosaur found with feathers. Now its tail had colors: reddish-brown rings alternating with pale ones, a little like a raccoon's tail.
That same year, another paper appeared in Science. Its subject was Anchiornis, a pigeon-sized dinosaur that had lived in what is now Liaoning around 160 million years ago. This paper became the template for almost every dinosaur color reconstruction that followed, and it remains the clearest way into the evidence chain. The researchers sampled the preserved head, torso, forelimbs, and hindlimbs, covering different feather types. They took samples at twenty-nine points, and every sample went into the scanning electron microscope.
The bodies they measured ranged from 100 nanometers to nearly 2,000 nanometers in length. Slender eumelanosomes predominated in most areas. At the far ends of the crest feathers were nearly spherical bodies about 500 nanometers across, corresponding to pheomelanin.
First, the researchers used black, gray, and brown feathers from living birds, whose colors were already known, to teach a statistical method how to classify them. These feathers formed the training set. They then used variation in the long axis, skewness in the short axis, aspect ratio, and density to assign the fossil bodies to known categories. This step is called discriminant analysis.
The result was a sharply patterned little dinosaur. Its body was mostly gray. The long feathers on its wings and legs were white with black tips, creating rows of black-and-white markings. A reddish-brown crest stood on its head, and reddish-brown spots dotted its face. The researchers compared its wing and leg patterns to those of a modern domestic chicken, the Spangled Hamburg. For the first time, a dinosaur had been colored from head to toe.
We need to pause here, because the next sentence is the most important one in this essay. Twenty-nine points are not twenty-nine pixels. A phone photograph contains more than ten million pixels.
The route from twenty-nine points to a continuous, full-body color image goes like this. Sample a point. Measure large numbers of little bodies at that point. Assign the point to a color category. Identify the feather region it belongs to, then extend that category to neighboring, unsampled feathers of the same kind. Finally, an artist joins the regions together. The picture you see online is a continuous field of color. The original measurements came from twenty-nine separate locations.
The reliability of the classification was uneven too. The paper stated that twenty-four of the twenty-nine samples had a classification probability above 0.9. Three had probabilities of only 0.56 to 0.74: clearly ambiguous assignments.
Several other locations contained so few melanosomes that they were interpreted as lacking pigment deposition and therefore as “light” or “nearly white.” Sample 14 was also a conspicuous outlier in the discriminant space.
Black and brown were relatively well separated by morphology. Gray was harder. Many combinations of shape and density can produce gray, so it occupied a broader region of the discriminant space and lay closer to other categories. The authors also explicitly acknowledged that pigments such as carotenoids and porphyrins leave no morphology that can be read in the same way. Their analysis could not rule out a contribution from these pigments to the feathers' actual colors.
There is another fact you almost never see beside a reconstruction: this Anchiornis specimen did not preserve a tail at all. The tail in the paper's illustration was filled in using another, more complete Anchiornis. This exemplary “whole-body reconstruction” thus combines information from three different sources: measurements of this specimen, anatomical completion from another specimen, and the artist's work of joining them together.
Finally, consider the scale of the calibration. The original training set for the Anchiornis study distinguished only three categories: black, brown, and gray.
When the team later reviewed the method, they reported roughly 90 percent classification accuracy on this restricted, three-way-choice problem. Three categories, 90 percent. That was the full weight the original calibration chain could bear.
The databases did expand. One comparative database commonly used in studies of dinosaurs and ancient birds grew to contain 168 feather samples from 101 living bird species.
A later study covered 181 living amniote taxa. Its most important result was not that “the mapping is more universal.” Quite the reverse: it found that a strong association between shape and color was meaningful in bird feathers and mammal hair, but could not simply be applied to the skin of lizards, turtles, and crocodilians.
In 2012, researchers added iridescent feathers and feathers of the kind found in penguins. The three-way choice became a five-way choice, and the accuracy of predicting feather color from melanosome shape alone fell to about 82 percent. In 2019, non-iridescent structural colors were added, and accuracy dropped to about 62 percent. Gray and these structural colors were particularly difficult to tell apart. Shape and color are not linked by a one-to-one lookup table. Their relationship is probabilistic, and the probability depends on the taxa used for training, the tissue, the way color categories are defined, and the statistical model.
III Melanosomes or Bacteria?
In 2014, seven researchers published a paper with a very direct title: Melanosomes or Microbes.
Its claim was narrow, which made it harder to answer: microscope images alone cannot rule out microorganisms.
The experiment was a comparison. Using scanning, field-emission scanning, and transmission electron microscopy, the researchers compared genuine melanosomes inside modern chicken feathers with microorganisms growing on feather surfaces and with the slimy film those microorganisms formed, like the slippery coating inside a water pipe.
They then reexamined a fossil feather assigned to Gansus yumenensis, along with a set of previously published fossil micrographs. Their conclusion came in three parts: sizes overlap, shapes overlap, and distributions can overlap too.
Size is the easiest part to understand. Melanosomes in bird feathers usually range from a few hundred nanometers to about two micrometers. Many spherical and rod-shaped bacteria fall within the same order of magnitude.
Later work on fossil frogs supplied an even more tangible example: some rod-shaped bacteria can grow to 1.5–2.5 micrometers. And with spherical or ellipsoidal bodies, scanning electron microscopy alone may not even distinguish a coccus from a melanosome. “About one micrometer” therefore has no diagnostic force. “A long rod” has only limited force. “A sphere” is especially dangerous.
Shape is not a binary switch either. Early popular-science illustrations often drew eumelanosomes as sausages and pheomelanosomes as meatballs. It works well for teaching, but is too crude for identification: bacteria themselves come as spheres, rods, and all sorts of forms in between. Melanosome aspect ratios also vary continuously, and differ from one tissue to another.
The advocates of the melanosome interpretation later changed tack. Instead of matching bodies one by one, they compared the statistical distributions of entire populations. They argued that the bodies in fossil feathers generally fell within the relatively narrow size and shape range of modern melanosomes, whereas the microorganisms in the control experiments spanned a much broader range of forms. By then, the argument itself had changed. The question was no longer “Does any bacterium look like a melanosome?” It was “Which kind of structure does this whole population most closely resemble?”
Two other kinds of evidence did the most to tip the balance. The first was arrangement. In the Crato feather, the bodies were concentrated in the original dark bands and aligned along the axes of the barbs and barbules.
The light bands contained almost none of the same carbonaceous bodies. In modern feathers, too, melanosomes enter keratin cells in accordance with the feather's developmental structure. They do not simply spread across the surface at random.
In the iridescent Messel feather, the structure was more specific still. The bodies formed a dense, continuous layer at the surface of the barbules, with a looser arrangement inside, matching the thin-film interference structures of living bird feathers. Asking a random decay biofilm to reproduce anatomical boundaries, directional alignment, and the layering required for the optics all at once places a much greater burden on that explanation. But this test has limits. Specimens that have been severely flattened, have lost all their keratin, and retain only a layer of microbodies are precisely the ones that lose this topological information. The discriminating power of an exceptionally preserved case cannot automatically be transferred to every fossil.
The second kind of evidence was location. A 2016 study of feathers from Eoconfuciusornis, an ancient bird from 130 million years ago, reported that the bodies stood in the expected relationship to preserved keratin-like structures within the original feather tissue. Molecular evidence of both keratin and melanin was recovered as well.
This directly addressed the central question raised by the 2014 paper. Modern melanosomes belong inside feather tissue; they are not a foreign bacterial film spread over it later. It did not prove that every previously published microbody was a melanosome. It did, however, substantially strengthen the general proposition that melanosomes really can survive in place within feathers.
Chemistry offers a third route, and the rules governing it need the clearest explanation. In 2012, researchers used a form of mass spectrometry that reads molecular fragments from a surface to analyze melanosome-like structures in the eyes of Eocene fish. The fragment spectra matched those of modern melanin, and the signals occupied the same locations as the bodies. In 2015, another group added diagenetic experiments to the same line of reasoning: after modern melanin underwent maturation at 200 and 250 degrees Celsius, its spectra moved progressively closer to those of the fossils.
Thus, “the fossil spectrum differs from fresh melanin” does not necessarily disprove an original melanin source. The difference can be the result of a repeatable diagenetic transformation. But identification relies on the statistical combination of a set of fragments. No single signal belongs exclusively to melanin.
The same applies to synchrotron elemental imaging. In 2011, researchers mapped copper and other trace elements in fossil feathers. They found copper bound to organic matter concentrated in regions that had originally contained melanin.
The method's strength is that it can map a large area, showing directly whether a pattern has chemically distinct regions. Its weakness is that copper, zinc, and calcium are not exclusive to melanin. Sediment and microorganisms can adsorb these elements, and microorganisms can produce melanin themselves.
Raman spectroscopy illuminates a sample with a laser and reads molecular vibrations. Its position is the trickiest of all: melanin and many disordered carbonaceous materials produce almost identical spectra. It can therefore say, “There is carbonaceous material here, consistent with melanin.” It cannot say, “These granules are melanosomes.”
Taken together, the three chemical approaches yield an unappealing but accurate sentence: detecting a melanin-like signal does not make every little body in the photograph a melanosome. The signal must also be shown to coincide spatially with the bodies. Sediment, microbial melanin, and other carbonaceous sources must still be excluded.
So where does that leave us? The answer depends on the case. Where preservation is excellent, the bodies follow the feather's barb structure and color bands, and the chemical signals coincide with them, bacteria have become a very weak alternative explanation. For isolated bodies exposed on a rock surface, with neither a tissue position nor chemical verification, bacteria remain a source that must be ruled out.
A 2020 methodological review proposed a step-by-step framework for checking the evidence. In practice, that is what it means: test each specimen on its own merits; issue no blanket license. This is a demanding rule. It withdraws the general permission implied by “such-and-such a method has proved these are melanosomes” and puts the burden of proof back on every individual specimen.
IV Shapes Change
Suppose we accept that the little bodies really are melanosomes. The argument has not ended; it has moved to another level. The question in the second round is how far we can travel from shape to color, even if the identity is right. There are five independent problems here. Solving one does not solve the next.
The first is that diagenesis changes shape. In 2013, five researchers subjected modern feathers to high temperatures and pressures to simulate maturation during burial. The conditions included 200 and 250 degrees Celsius, at a pressure of about 250 atmospheres.
The melanosomes were not destroyed. Quite the contrary: they persisted after other feather structures had visibly degraded. But their size, shape, and spatial distribution changed. Some samples showed linear shrinkage approaching 20 percent.
This struck directly at a hidden assumption in the classification chain. The modern training set uses lengths, widths, and aspect ratios that have not undergone diagenesis. The fossil data fed into the classification model are lengths, widths, and aspect ratios after an unknown thermal history. The two sides of the comparison are not the same thing.
Nor can we say with certainty how much they change. In another set of experiments in 2015, melanosomes shrank roughly proportionally by 10 to 20 percent. Experimental conditions are specified; natural fossils each have their own burial history. We know that shape changes, but not how much it changed in each particular specimen.
Experiments in 2023 went a step further. Modern melanin-bearing feathers were embedded in compacted sediment and matured under conditions closer to natural pore systems, at temperatures ranging from 190 to 300 degrees Celsius. Between 190 and 225 degrees, feathers of different colors gradually became much the same brown stain. Black, reddish-brown: they all ended up brown.
Near 250 degrees, carbonization intensified. Near 300 degrees, more organic carbon was lost. One detail will matter later: at high levels of maturation, eumelanin appears to be more stable than pheomelanin.
There is therefore no formula at present that can work backward from a fossil's dimensions to its dimensions in life. A more dependable approach is to compare differently colored regions within the same specimen, rather than comparing them directly with fresh modern feathers.
The second problem is more fundamental: even if shape survives with perfect fidelity, shape alone may not specify a unique color.
As Section 2 explained, the relationship works reasonably well in bird feathers, but breaks down in the skin of lizards, turtles, and crocodilians. What about mammal hair?
A large-sample study in 2025 put the problem more starkly. The researchers measured more than 7,600 melanosomes from fifty-nine living mammal species, plus a Jurassic fossil. Melanosome morphology explained only about a quarter of the variation in “redness,” and less than a tenth of the variation in lightness.
The most vivid example was this: African and Asian elephants have different body colors, but almost identical melanosome geometry.
Conversely, different combinations can produce the same color. Between a granule's shape and a color the human eye can name lie the pigment's chemical composition, the number and density of the granules, the way they are embedded in keratin, and the scattering of light. Early models generally considered only two or three of these.
The third problem is structural color. Iridescence is not the color of a pigment itself. It comes from nanoscale layers of materials with different refractive indices. The colors on a soap bubble work the same way.
In feathers, interfaces between melanosomes, keratin, and air can form thin films or multilayer structures. Light reflected from neighboring interfaces interferes, strengthening certain wavelengths. Change the layer thickness, the spacing of the melanosomes, or the angle of incoming light, and the peak color changes.
Finding melanosomes is therefore not enough to establish iridescence in a fossil. You also need to see their layered arrangement and measure the distances between the layers. Those are precisely the dimensions compaction changes. A shift of a few nanometers or a few tens of nanometers may change the apparent color.
Microraptor is the classic example. A 2012 study described it as black all over, with a blue-black sheen like a crow's feathers. The black is on relatively firm ground: there are abundant melanosomes, and their shapes fall within the range found in deep-black feathers.
The sheen is less secure. The original keratin is usually gone, and a flattened fossil need not retain its original layer spacing. “Very black, possibly glossy” is therefore much more reliable than “this patch had exactly this blue-black metallic sheen.”
The fourth problem is the method's own blind spot. Carotenoids produce many yellows, oranges, and reds. Flamingo pink and canary yellow both depend on them. Porphyrins can produce reds, greens, and other hues.
These molecules do not form granules whose shape can be read, a limitation acknowledged in the original Anchiornis paper itself. They are not wholly incapable of surviving, but they are much harder to identify than melanin. Even when a chemical marker is detected, locating it in the original animal is difficult.
The fifth problem: melanosomes do not belong exclusively to feathers and skin. In living vertebrates, the retinal pigment epithelium, pigment cells in the skin, and some internal organs all contain large numbers of them. We ourselves have a layer of melanin at the backs of our eyes, absorbing excess light like the dark interior of a camera.
Studies of living and fossil frogs found that internal tissues such as the liver, lungs, and spleen can contain even more melanosomes than the skin. Decay experiments show that most internal-organ melanosomes remain in the torso, but disturbance, decay, and flowing water can carry some away from their original positions. In one survey of seventy-three fossil frogs, about 93 percent preserved melanosomes from tissues other than the body surface. In roughly a quarter of those cases, their distribution extended from the torso into the limbs.
A layer of black bodies covering the outline of an animal is therefore not necessarily “the original skin color.” It may include melanosomes redistributed from the eyes, liver, peritoneum, or other tissues after they broke down.
“Identifying melanosomes” thus really means answering two separate questions: Is this a melanosome? And which tissue did this melanosome originally belong to? Chemistry can help with the first. The second still depends chiefly on anatomical location.
V The Pictures That Had to Change
The best test of an evidence chain's reliability is not what it claims, but what has been revised, and at which level.
Sinosauropteryx has undergone some of the largest revisions. The famous 2010 conclusion identified light and dark tail rings and interpreted some melanosome shapes as chestnut or reddish-brown. In 2017, researchers systematically remapped several Sinosauropteryx specimens. They did more than confirm the tail bands: they proposed a dark back, a pale belly, a dark “mask” around the eyes like a masked bandit, and pronounced countershading.
The direction of the revision is what matters. The pattern expanded, from tail bands to a whole-body back-to-belly gradient and a mask. Certainty about hue, however, contracted. As problems with pheomelanin preservation and corrections for shape became apparent, “light and dark areas with different melanin distributions” became a much safer claim than insisting on one exact shade of chestnut. A picture can become richer and more cautious at the same time. There is no contradiction.
Gansus yumenensis was an ancient bird that lived near water more than 100 million years ago. Its fossils were found at Yumen, in Gansu. Here the revision came at a different, deeper level. Early work had included this specimen's feather microbodies in the evidence for melanosomes.
In 2014, Melanosomes or Microbes reexamined the feather and concluded that the bodies seen under a scanning electron microscope were not enough to establish their identity. What changed was not a particular hue in a “color portrait of Gansus.” It was whether the object supporting color reconstruction had been correctly identified at all.
Archaeopteryx, often treated as the earliest bird, lived around 150 million years ago in what is now Germany. The story of its isolated feather is more tangled. Unearthed in Germany in 1860 or 1861, it was described in 1861, even before the Archaeopteryx skeleton. The name Archaeopteryx was first given to this feather.
A 2012 study interpreted the celebrated feather as heavily melanized and nearly black, bringing it into the discussion of Archaeopteryx plumage. In 2013, another group used synchrotron chemical imaging and argued that heavy pigmentation was concentrated mainly at the tip and in the outer vane.
In 2019, researchers used laser-stimulated fluorescence to recover the outline of the quill, which had been removed during early preparation. They argued that the feather matched no feather type known on Archaeopteryx and might not belong to it at all. In 2020, another group made a renewed case that the feather was consistent with an Archaeopteryx identity.
Through this dispute, the local chain from “little bodies” to “melanin” to “black” remained largely untouched. What changed was whether the black feather belonged to Archaeopteryx, where on the animal it had grown, and whether it could stand for the whole animal. A correct measurement does not guarantee correct placement in a reconstruction. Improving color measurement alone cannot establish ownership or anatomical placement; that needs other evidence, including anatomy and taxonomy.
Borealopelta was an ankylosaurian dinosaur, one of the armored dinosaurs, encountered by an excavator operator in a Canadian oil-sands mine in 2011. More than five meters long and weighing over a metric ton, it was preserved in three dimensions, skin and armor together, like a stone sculpture.
A 2017 paper reconstructed a dark-back, pale-belly pattern from organic traces on its skin and suggested a reddish-brown coloration. If an animal this large still needed countershading to hide, the predators eating it must have been formidable. It also makes a good example of “pattern is more reliable than color name.”
But broader chemical work found that sulfur-containing aromatic compounds and signals from diagenetically altered melanin can be affected by maturation, sulfurization, and other organic matter. A “spatial gradient of countershading” and “precisely reddish-brown” are therefore not conclusions of equal strength. The uniform coat of brown in a public reconstruction visually compresses them into a single fact.
The most elegant experiment was carried out on Psittacosaurus, a dog-sized plant-eating dinosaur with a parrot-like beak. Dark backs and pale bellies are common in living animals. The pattern is called countershading: light shines down from above, so the belly is naturally in shadow. Darken the back and lighten the belly, and the animal looks flatter, making it harder to spot.
The researchers first mapped the distribution of pigment across Psittacosaurus's back and belly from preserved skin traces. They then built a life-size body model to the dimensions of the skeleton, painted it a uniform gray, and placed it under two kinds of light: diffuse light beneath a forest canopy and direct sunlight in the open.
They measured the shadows the model cast on itself, calculated the dark-to-light transition that would be needed to cancel those shadows, and compared the predicted boundary with the actual boundary on the fossil. The match was closer under diffuse forest light. They therefore inferred that Psittacosaurus had probably lived in woodland.
Each link in this chain is less firm than the one before. The light-and-dark distribution on the skin is visible, and that is the firmest link. Calling it camouflage is a functional interpretation, less secure. Matching it to forest light relies on an optical model.
Saying the animal lived in a forest adds assumptions about how it usually stood and moved. That is the least secure link. The 2017 conclusion that Sinosauropteryx “lived in open habitats” used the same logic. The last sentence cannot be written with the same certainty as the first.
VI What in That Picture Was Measured?
We can now answer the question posed at the beginning: how much of a published dinosaur color reconstruction is supported by measurement?
Anchiornis gives the clearest example. The evidence corresponds directly to twenty-nine sampling locations and the feather regions around them. Twenty-four had classification probabilities above 0.9; several were markedly lower. Getting from those points to a continuous, whole-body field of color requires interpolation. The tail was not even preserved in this specimen.
A reconstruction can therefore be divided into four layers. The direct-measurement layer: the size, shape, density, and statistical classification of bodies at the sampled locations. The spatial-extrapolation layer: unsampled parts of the same feather region are assumed to match the sample point.
The anatomical-completion layer: feather outlines missing from this fossil but inferred from other specimens. The artistic layer: continuous gradients, exact saturation, the intensity of sheen, eye color, the color inside the mouth, every unpreserved soft tissue, and the lighting and background.
These four layers occupy very different amounts of the picture, and the image does not tell you where their boundaries lie. It is a little like an old photograph that has been colorized. The photograph is real; the colors were added afterward, stroke by stroke, and the viewer has trouble telling which strokes have a basis in evidence. A picture without a key spreads three different strengths of knowledge across one continuous field of color. What reaches the reader is visually uniform throughout.
One color needs a discussion of its own: white.
The Anchiornis study interpreted feather bases with very few melanosomes as pale, close to white. As an analogy with modern bird feathers, that is a reasonable inference.
Most white bird feathers today are white because they lack pigment. Light scatters in many directions through the keratin, and the feather looks white. Many white gulls nevertheless have black wingtips: melanin makes feathers more resistant to wear, so the hardest-working parts are black.
Logically, though, “no melanosomes” allows at least four possibilities. The region really was unpigmented and white. It contained non-melanin pigments, such as carotenoids. It had structural color that did not depend on melanosomes. Or the melanosomes and tissue were lost during preservation.
An asymmetry follows: “black” can often have positive evidence, while “white” often depends partly on negative evidence. They should not be assigned the same evidential rank, yet in a reconstruction they look equally certain.
Here the rule from Essay 11 applies: a failure to find something is informative only to the extent that we would expect to detect it if it were there. The relevant question is this: if that feather region had originally been rich in melanin, how likely would the same preservation conditions and methods have been to leave us detectable melanosomes?
Only if that probability is high does the blank count against an originally melanin-rich feather. It still does not establish white on its own or rule out the other possibilities—especially original carotenoids, which this morphological method cannot see in the first place.
Public images also merge two other things: a color category and an artist's chosen color value.
Discriminant analysis usually returns broad categories such as “black,” “gray,” “brown,” and “reddish-brown.” It does not give a single exact value on a palette. Within one category, real animals may vary greatly in brightness, saturation, and sheen. The particular gray, chestnut red, or blue-black in a paper's illustration should therefore be understood as “one visual example consistent with the paper's color category,” rather than the one and only color code an instrument extracted from the rock.
Is there a rule an ordinary reader can use? There is. When looking at a dinosaur reconstruction, separate “pattern” from “color name.” Light and dark bands, spots, masks around the eyes, back-to-belly gradients: those are patterns.
If the spatial arrangement is well preserved, evidence for a pattern can be quite strong. Try it with Anchiornis: the black-and-white markings on its wings and the crest on its head belong to pattern. Whether that crest was “actually chestnut or reddish-brown” belongs to color naming. That claim must pass three tests: correction for diagenesis, taxon-specific mapping, and density. Its evidential rank is distinctly lower.
The rule has a useful side effect: it explains why revisions always change color names first and patterns later. The Sinosauropteryx revision had exactly that shape. The pattern expanded while the color claim narrowed. The same new evidence can push the two levels in opposite directions.
This is not an unreasonable demand. Recovering whole-body color requires extensive sampling, and that sampling is often destructive.
A large specimen cannot be put into an electron microscope whole. Taking a sample often means scraping a small flake off the fossil. Real research is necessarily limited by sampling density, and limited sampling may not capture variation across the whole body. That limitation belongs beside the reconstruction, not buried in supplementary methods.
The difficulty is that a finished picture can make unfinished evidence disappear from view. Images can show uncertainty—with legends, blank areas, or alternative versions. Without such signals, though, we easily mistake a clear depiction for clear knowledge.
VII Where the Same Road Has Led
Ancient birds are the most natural extension of this method, since its calibration came mainly from modern bird feathers in the first place. Confuciusornis, Eoconfuciusornis, several enantiornithines, and feathers from Messel have all been used to study melanosome morphology, feather patterns, and structural color.
Eoconfuciusornis is especially important. Evidence that melanosomes and keratin survived together directly strengthens the interpretation that the bodies originated within the animal. Reliability usually turns on a very simple question: can the positions of the barbs and barbules still be recognized? Isolated bodies with no clear place in feather anatomy are distinctly less reliable.
Mammaliaforms are currently the most interesting direction, because a successful extension and a warning about calibration arrived in the same year. In 2025, researchers built a hair-melanosome database from 116 modern mammal species and reconstructed six Mesozoic mammaliaform fossils dating from roughly 165 million to 120 million years ago. Mammaliaforms include mammals and their closest ancient relatives, most of them only mouse-sized at the time. The reconstructions produced remarkably similar results overall: uniformly dark-brown coats.
That same year, another study of more than 7,600 melanosomes from fifty-nine living mammal species found a generally weak association between shape and fur color, with a pronounced phylogenetic effect. These results belong side by side. We should not choose one and discard the other.
The encouraging result is that useful signals can be extracted when the training set is specifically mammalian and the predictions are restricted to broad color categories. The limiting result is that the relationship between shape and color in mammal hair is far from one-to-one. Mesozoic fur colors are therefore best understood as probabilistic predictions. The simple early rule for bird feathers, “sausages versus meatballs,” should not be carried over.
Marine reptiles took the method from feathers into skin. Researchers reported eumelanin and melanosome-like structures in skin traces from ichthyosaurs, mosasaurs, and an ancient leatherback turtle. They proposed that at least some ichthyosaurs were dark across their body surfaces, like living leatherbacks, and drew comparisons with camouflage, ultraviolet protection, or thermal physiology in modern marine animals. A 2018 study of another Jurassic ichthyosaur, however, concluded that it had a dark back and a pale belly.
But the relationship between shape and color is less stable in skin than in bird feathers, and the movement of internal-organ melanosomes is an especially serious problem in soft tissues other than feathers. Here too there are two levels of strength. “There is melanin in the skin region” can be a relatively strong claim. “Shape tells us exactly which brown this patch of skin was” is much weaker.
Fossil insects follow another branch. Many colors in insect wing cases and wing scales are produced by nanoscale chitin layers or photonic structures; that is how some living beetles get their metallic green. Original reflection can be calculated from preserved layer spacing, though diagenesis can shift it.
The road opened by “dinosaur color” has thus split into two branches: proxies based on organelles and pigment chemistry, and proxies based on nanoscale optical structures. The branches sometimes meet in bird feathers, because melanosomes themselves can form optical structures.
How far back in time can this go? For a long time, the oldest widely cited record came from the Late Carboniferous, around 307 million years ago. In fossils from Mazon Creek, Illinois, in the United States, researchers identified melanosome-like structures in the eyes of jawless fish such as lampreys and hagfish, and on the body surfaces of lampreys.
The Tully monster discussed in Essay 13 came from the same place. Identification rested on size, shape, mode of preservation, and the restriction of the bodies to their original anatomical structures.
In 2026, the record leapt back to the Cambrian, about 518 million years ago. In some of the earliest vertebrate fossils from Chengjiang, Yunnan, researchers found melanosomes containing melanin, just like those in the layer at the backs of our eyes, and detected melanin's chemical signal as well. Those animals had four eyes: one pair on the sides of the head and two more on top.
Neither example amounts to a full-body color portrait of an ancient animal. Most of the evidence comes from tissues such as eyes, which are heavily melanized to begin with. They tell us more about “how long melanin can survive” than about “how finely we can reconstruct whole-body color.” News headlines have repeatedly merged those two questions.
One further kind of extension demands particular care: this approach has been used to answer questions beyond color. Melanosome shape and distribution have been enlisted in discussions of physiological change, the evolution of metabolism, feather evolution, and even the phylogenetic positions of fossil animals. The large 2014 study went so far as to suggest that diversification of melanosomes in feathered dinosaurs might be associated with broader physiological changes.
The problem is that the same study demonstrated how strongly melanosome morphology depends on tissue and taxon. Treating “similar melanosome shapes” directly as a strong character for establishing relationships between distant groups risks circularity. You infer relationships from a quantity known to vary with tissue and group, while the rules governing its variation have themselves been read from those relationships. The treatment of the Late Carboniferous material was more cautious by comparison: it used anatomical position and the layering of pigment cells, rather than treating the aspect ratio of an individual body as a phylogenetic code.
A more defensible boundary is a descending scale of what the evidence permits us to say. From melanosomes to pigment category: measurable in the right tissue, with several lines of evidence.
From pigment category to visible color: probabilistic. From visible color to function: open to multiple explanations. From function to behavior and environment: model-dependent. From behavior to evolutionary cause: color alone cannot decide. With every step, what can be said becomes weaker, while public accounts usually repeat only the final step.
VIII After a Remainder Is Drawn In
We can now return to what this essay is really about. What dinosaurs looked like is one half of it. The other half is what happens after something once impossible to know becomes measurable. The chain that reads its way from little bodies to color was built by people so that they could judge and compare. This series calls such things constructs.
First, let us establish that the old verdict really has been revoked. We can now say with reasonable confidence that some well-preserved fossil feathers, skin, and eye tissues retain original melanin or melanosomes, or their diagenetic remnants.
Some specimens have genuine light and dark bands, spots, masks, or back-to-belly pigment gradients. The rings on the tail of Sinosauropteryx, the black-and-white markings on the wings of Anchiornis, and the dark backs and pale bellies of Psittacosaurus and Borealopelta all belong here. In well-preserved, bird-feather-type tissues, very slender, densely packed melanosomes usually support heavily eumelanized black or dark-brown regions. These are not small conclusions. Twenty years ago, every one of them sat in the column marked “unknowable in principle.”
But the manner of that revocation matters. It is not that “color can finally be measured.” What was once unrecoverable as a whole has been broken down into a series of steps, each measurable in its own right, each with its own error.
Once it was broken apart, new remainders appeared immediately, and there was more than one. This series calls a remainder the part a construct cannot accommodate and cannot eliminate. How much change did diagenesis cause? Every fossil is different; there is no universal constant. The mapping from shape to color depends on taxon and tissue: usable in bird feathers, unusable in crocodilian skin.
Color-producing mechanisms beyond melanin are a systematic blind spot: many yellows, oranges, and vivid reds come from other pigments, while many blues come from other nanostructures. Melanosome shape alone cannot recover them. The bodies may come from other organs: a black layer need not be black skin. Sampling density has a ceiling: a large specimen cannot go into an electron microscope whole.
There is a point here that is easy to skip: these new remainders are not fragments of the old remainder.
The old remainder, “color cannot be recovered,” did not contain the question “By what percentage did diagenesis shrink it?” It could not have contained it. Without a measurable object, there can be no question about that object's deformation.
Likewise, “the mapping from shape to color fails in crocodilian skin” could not have been said before that mapping existed. As the construct opens out another level, the remainder opens out another level too. New remainders are products of a new construct, not debris left behind because the old remainder was inadequately cleared away.
The remainder here is produced by the very act that turns the unmeasurable into the measurable. It grows alongside our capabilities, becoming more specific as they grow stronger. We should not assume, then, that slightly better methods will make these remainders disappear.
There is also a question of direction, and it runs deeper than any of the problems above.
What this method sees most clearly is black, dark brown, some grays, some reddish-browns, and structural sheen that melanosome arrays help produce. What it most readily misses is carotenoid yellow, orange, and vivid red; poorly preserved non-melanin molecules; blue and ultraviolet colors produced by keratin-and-air nanostructures in which melanosomes are not the main component; and greens produced by a mixture of mechanisms. Add the 2023 finding: at high levels of maturation, eumelanin appears more stable than pheomelanin.
Put the two together and the conclusion is this: the Mesozoic world of color we are reconstructing is biased toward black and dark brown at the level of the method itself. If a dinosaur was bright yellow because of other pigments, this method could miss the yellow. That would not give us grounds to paint it brown.
Essay 11 dealt with preservation bias: what can become a fossil. The bias here comes one step later. The remains already exist, and they genuinely carry information. The bias lies in the measurement proxy itself.
The properties visible to that proxy and the properties that actually existed are not the same set. A conclusion such as “Mesozoic mammals were generally dark brown” therefore needs to be read twice. The first time, read what it says. The second time, ask: if some of them had been bright yellow, would this method have seen them?
What we have here is the full course of a remainder being drawn into a construct.
Put these twenty years in order and the shape emerges. At first, there was no measurable proxy. A proxy was found. It turned out the proxy might have been misidentified. Chemical and topological checks were added.
The proxy's identity was partly accepted. Its measured values were found to change during diagenesis. The modern mapping itself was found to depend on taxon and tissue. Every step closed one gap and opened another.
This does not push color back into the “unknowable.” Quite the opposite. We can say far more about dinosaur color today than we could twenty years ago, and every statement comes with sources of error that can be traced. Traceability is itself a gain: it turns disagreements into things we can argue about, instead of people talking past one another.
Yet the picture remains open. Of its four layers, only the first is measured. An entire region of its color space lies outside the method. Every new conclusion leaves a new question beside it.
We can keep painting. With each stroke, though, remember which colors came from the fossil and which we have lent it.