Many wildlife species are propagated in captivity as models for behavioral, physiological, and genetic research or to provide assurance populations to protect threatened species. However, very little is known about how animals evolve in the novel environment of captivity. The histories of most laboratory strains are poorly documented, and protected populations of wildlife species are usually too small and too short-term to allow robust statistical analysis. To document the evolutionary change in captive breeding programs, we monitored reproduction and behavior across 18 generations in six experimental populations of
Peromyscus
leucopus
mice started from a common set of 20 wild-caught founders. The mice were propagated under three breeding protocols: a strategy to retain maximal genetic diversity, artificial selection against stereotypic behaviors that were hypothesized to reflect poor adaptation to captivity, and random bred controls. Two replicates were maintained with each protocol, and inter-replicate crosses at generations 19 and 20 were used to reverse accumulated inbreeding. We found that one of the stereotypic behaviors (repetitive flipping) was positively associated with reproductive fitness, while the other (gnawing) was relatively invariant. Selection to reduce these stereotypic behaviors caused marked reduction in reproduction, and populations not under artificial selection to reduce these behaviors responded with large increases in flipping. In non-selected populations, there was rapid evolution toward much higher proportion of pairs breeding and more rapid conception. Litter size, pup survival, and weaning mass all declined slowly, to the extent that would be predicted based on inbreeding depression. Inter-crossing between replicate populations reversed these declines in fitness components but did not reverse the changes in behavior or the accelerated breeding. These findings indicate that adaptation to captivity can be rapid, affecting reproductive patterns and behaviors, even under breeding protocols designed to minimize the rate of genetic change due to random drift and inadvertent selection.