Evolution Icon Evolution

Why Fossils Cannot Demonstrate Darwinian Evolution

William A. Dembski


Editor’s note: William Dembski and Jonathan Wells, leading figures in the intelligent design movement, are co-authors of The Design of Life: Discovering Signs of Intelligence in Biological Systems. Originally published by the Foundation for Thought and Ethics, this path-breaking work explores some of the most important arguments for intelligent design in biology. To celebrate the launch of Foundation for Thought & Ethics Books as an imprint of Discovery Institute Press, we have been publishing excerpts from the book here at Evolution News. Through this Friday, July 8, we will be making The Design of Life available for only $10 — more than a 70 percent discount. That includes both the full-color hardcover and an accompanying CD with additional materials. If you haven’t read this classic book, now is your chance! Order now, because this special discount concludes at the end of this week.

The transition from reptiles into mammals via mammal-like reptiles is regarded by many evolutionary theorists as the best example of an evolutionary lineage in the fossil record. There are, however, three fundamental problems with this and all other examples of inferring Darwinian evolution on the basis of fossil evidence. The first is that any specific hypothesis must use the fossil data selectively; the second is that similarities in fossil or living organisms may not be due to common ancestry; and the third is that fossils cannot, in principle, establish biological relationships.

DOL COVER special offer.jpg(1) Using the fossil evidence selectively. As in the case of therapsids, fossils more mammal-like can occur earlier in the fossil record than fossils that are less mammal-like. Yet to trace an evolutionary lineage on the basis of the fossil record requires that therapsids structurally more similar to mammals enter the history of life later than those that are structurally less similar. Evolution, after all, needs to follow time’s arrow and cannot have offspring giving birth to parents.

A similar problem arises with geographical mismatches, in which fossil organisms that are supposedly next to each other in a structural progression are widely separated geographically. If the geographical separation is too great, how can one organism be ancestral to the other? Reproduction, after all, requires proximity — parents do not give birth to offspring at the other side of the globe.

The problem of temporal and geographical mismatches is widespread. The Darwinist’s way around this problem is to assume that organisms that appear to enter the fossil record too late or too far away actually entered earlier or closer together. But such assumptions are entirely ad hoc and ignore the actual fossil evidence.

This illustrates a larger problem — what scientists call “cherry-picking.” Given a sufficiently large data set, it’s possible to find salient patterns simply by trying out enough different ways of combining items of data. Many structural progressions found in the fossil record are nothing more than “cherries” — in other words, they are statistical artifacts that result from trying out enough different ways of combining fossil data. The sheer quantity of fossil data is immense. Simply by combining and recombining these data in enough different ways and by attending to sufficiently many distinct features of structural similarity, it is possible to generate reasonably long fossil progressions arranged by structural similarity.

Two well-known results from statistics give rise to the cherry-picking fallacy. One is the birthday paradox. Although there are 365 days in the year, it only takes 23 people, chosen at random, for there to be a better than even chance that at least two of them share a birthday.1 That’s because in calculating the probability of a shared birthday, we must factor in all possible ways of pairing these 23 people. As it turns out, there are 253 pairings and thus 253 ways that any two of them might share a birthday (it’s not coincidental that 253 is over half of 365; that’s why 23 people are more likely than not to share a birthday). Because of the birthday paradox, the fossil record readily yields fossils that match up on a given feature of similarity quite apart from any underlying cause.

The other result from statistics that gives rise to the cherry-picking fallacy is the file-drawer effect. Suppose you claim that a coin you are flipping is biased because you just now flipped it ten times and each time it came up heads. The degree to which you are justified in claiming that the coin is biased will depend on the unreported number of times you flipped the coin before actually reporting ten heads in a row. The file-drawer effect refers to the unsuccessful studies that go unreported and languish in a researcher’s file-drawer.2 The bigger the file-drawer, the greater the number of unsuccessful studies that went unreported and, consequently, the less compelling is any eventual report of success. Even with a fair coin, after a few thousand coin flips, one is virtually assured of flipping ten heads in a row. Thus, if your file-drawer contains thousands of unreported coin flips, the ten heads in a row you report can’t confirm that the coin is biased.

Likewise, for every “successful” structural progression in the fossil record (like the reptile-to-mammal progression), there are all too many “unsuccessful” ones, conveniently unreported and languishing in evolutionary biology’s “file-drawer.” Evolutionary biology’s file-drawer of failed attempts at finding such fossil progressions is huge. For instance, where are the progressions based on structural similarity that connect the different animal phyla — progressions that should be there if evolutionary theory is correct? Despite a massive search of the fossil record by paleontologists and evolutionary biologists, no such progressions are known. In consequence, there is every reason to be suspicious of using “successful” fossil progressions to infer evolutionary lineages.

(2) Similarity may not be due to common ancestry. In evolutionary theory, convergence refers to the origination of identical or highly similar structures through independent evolutionary pathways rather than inheritance from a common ancestor. Darwinian theory attributes convergence to similar environments that apply similar selection pressures and thereby produce similar structures.

This explanation is on its face implausible because there is no reason to think that Darwin’s opportunistic mechanism has the fine discrimination to produce virtually identical complex structures in causally disconnected environments. Yet organisms possess many similar features not thought to arise from a common ancestor. Convergence is a widespread fact. As a result, even if Darwinian theory were true, one could never be sure whether similar features shared by two fossils resulted from convergence or from common ancestry. If similar structures can evolve and re-evolve repeatedly, then fossils cannot distinguish convergence from common ancestry, and tracing evolutionary lineages in the fossil record becomes impossible.

In fact, similarities may not be due to Darwinian evolution at all. In a 1990 book intended to refute critics of Darwinian evolution, biologist Tim Berra used pictures of various models of Corvette automobiles to illustrate how the fossil record provides evidence for descent with modification. “If you compare a 1953 and a 1954 Corvette, side by side,” he wrote, “then a 1954 and a 1955 model, and so on, the descent with modification is overwhelmingly obvious.”3 But automobiles are designed, not descended from other automobiles. Berra actually proved the opposite of what he intended, namely, that a series of similarities could be a product of intelligent design rather than Darwinian evolution.

The case for Darwinian evolution would be greatly strengthened if scientists could demonstrate (rather than merely gesture at) a plausible mechanism for producing macroevolution. But they have been unable to do so. Even if we assume that a structural progression such as the therapsid-to-mammal sequence is an evolutionary lineage, the fact remains that we know of no material mechanism capable of producing it. To be sure, one can tell a story about how a Darwinian mechanism might have caused the progression, but that’s all it would be — a fanciful story.

Take the evolution of the mammalian ear from the reptilian jaw. How exactly did those two bones from the reptilian jaw “migrate” to the mammalian ear? The word “migrate” in this context is empty of scientific content. What genetic changes and selection pressures were in fact operating, and how, specifically, did they bring about the evolutionary pathway in question? No such details are known. Yet, without such details, there is no way to assess whether the Darwinian mechanism was even capable of, much less responsible for, evolving the mammalian ear.

Perhaps a sufficiently adept designing intelligence could change the reptilian jaw into the mammalian ear. But an intelligently guided process would not be Darwinian.

(3) Fossils cannot, in principle, establish biological relationships. Imagine finding two human skeletons in the same location, one apparently about thirty years older than the other. Was the older individual the parent of the younger? Simply by looking at the skeletons, one can’t say. Without independent evidence (e.g., genealogical, dental, or molecular), it is impossible to answer the question. Yet in this case we’re dealing with two skeletons from the same species that are only a generation apart. It follows that even if we had a fossil representing every generation and every imaginable intermediate between, say, reptiles and mammals — if there were no missing links whatsoever — it would still be impossible, in principle, to establish ancestor-descendant relationships.

In 1978, fossil expert Gareth Nelson, of the American Museum of Natural History in New York, wrote: “The idea that one can go to the fossil record and expect to empirically recover an ancestor-descendant sequence, be it of species, genera, families, or whatever, has been, and continues to be, a pernicious illusion.”4

Henry Gee, a science writer for Nature, doesn’t doubt Darwinian evolution, but he likewise admits that we can’t infer descent with modification from fossils. “No fossil is buried with its birth certificate,” he wrote in 1999. “That, and the scarcity of fossils, means that it is effectively impossible to link fossils into chains of cause and effect in any valid way.” According to Gee, we call new fossil discoveries missing links “as if the chain of ancestry and descent were a real object for our contemplation, and not what it really is: a completely human invention created after the fact, shaped to accord with human prejudices.” He concluded: “To take a line of fossils and claim that they represent a lineage is not a scientific hypothesis that can be tested, but an assertion that carries the same validity as a bedtime story — amusing, perhaps even instructive, but not scientific.”5

In short, fossils cannot demonstrate Darwinian evolution.


(1) William Feller, An Introduction to Probability Theory and Its Applications, 3rd ed., vol. 1 (New York: Wiley, 1968), 33.

(2) S. Iyengar and J. Greenhouse, “Selection Models and the File Drawer Problem (with Discussion),” Statistical Science 3 (1988): 109-135.

(3) Tim Berra, Evolution and the Myth of Creationism (Stanford, Calif.: Stanford University Press, 1990), 117-119.

(4) From a presentation by Gareth Nelson in 1969 to the American Museum of Natural History, quoted in David M. Williams and Malte C. Ebach, “The Reform of Palaeontology and the Rise of Biogeography — 25 Years after ‘Ontogeny, Phylogeny, Palaeontology and the Biogenetic Law’ (Nelson, 1978),” Journal of Biogeography 31 (2004): 709.

(5) Henry Gee, In Search of Deep Time (New York: Free Press, 1999), 23, 32, 116-117.

Photo: Royal Ontario Museum, by Daderot [CC0 or CC0], via Wikimedia Commons.

William A. Dembski

Board of Directors, Discovery Institute
A mathematician and philosopher, Bill Dembski is the author/editor of more than 20 books as well as the writer of peer-reviewed articles spanning mathematics, engineering, philosophy, and theology. A past philosophy professor, he retired in 2014 from active research and teaching in intelligent design (ID) to focus on the connections between freedom, technology, and education — specifically, how education helps to advance human freedom with the aid of technology. Bill Dembski is presently an entrepreneur who builds educational software and websites. He lives near Denton, Texas.