You can train a dog to catch mice but not to meow. Training has its limits; development is only so malleable; the organism gives only so far, and then resists. What are the limits? How malleable? How far? The underlying problem should be familiar, and it is the same one that lies at the heart of an ongoing debate between Stephen Jay Gould and Mark Ridley (see Comment and Reply, this issue). The questions raised by the debate are different: Is morphology perfectly plastic to natural selection over evolutionary time, or does it resist, with an internal dynamic of its own? Is a species a passive particle, buffeted this way and that by external forces, or is it channeled by development? But it is still the naturenurture problem, considered now for entities far larger than organisms (lineages) and on time scales far longer than lifetimes (millions of years). The narrower, more immediate issue in this debate is the fascinating hypothesis presented by Gould in his 1989 book Wonderful Life, a hypothesis that may already be blossoming into a major research program in paleobiology. Gould's proposal is that the animals of the Burgess Shale are morphologically more disparate overall than a comparable group of modern animals, that Burgess animals are quite variable in the characters that today are quite stable and define higher-level taxa, and that both findings may be a consequence of the greater developmental flexibility of the ancient animals. He suggests that decreasing developmental flexibility, or increasing developmental entrenchment, may be a basic feature of the evolutionary process. (Gould's proposal encompassed all of the Metazoa, but the debate has focused on the arthropods, so I will concentrate on this subset of the problem also.) Ridley (1990) challenged Gould's idea; Gould (1991) then responded to Ridley and other critics. The exchange in this issue marks the second round. The debate has been lively and productive, but it has reached the point where little remains that can be resolved by argument alone. Gould and Ridley (this issue) seem to agree that direct empirical tests are now crucial. Still, a bit of conceptual sorting-out remains to be done. Ridley (1990, this issue) understands Gould (1989) to be claiming that the constancy of the characters defining arthropod higher taxa (mainly the pattern of head segmentation) is the evidence for their developmental stabilization. Ridley (1990) responds that some characters are likely to remain unchanged by chance, even if none is especially stabilized, and these will be the characters modern taxonomists use to define the taxa. Gould's argument does not work, in other words, because character constancy has another explanation. Gould (1991) agrees that the argument does not work, calling it the retrospective fallacy, and says he never made it. Ridley (this issue) further understands Gould's argument to rely on the low consistency index of cladograms for Cambrian arthropods as evidence for their developmental lability. He contends that the consistency index is not low, and if it had been low, it would not be evidence for lability because a low index has other possible explanations, such as high variability in selection in the Cam-