We tested whether selective breeding for early-age high voluntary exercise behavior over 16 generations caused the evolution of lifelong exercise behavior, life expectancy, and age-specific mortality in house mice (Mus domesticus). Sixteenth-generation mice from four replicate selection lines and four replicate random-bred control lines were individually housed from weaning through death and divided between two activity treatments (either with or without running wheels). Thus, there were four treatment groups: selection versus control crossed with active versus sedentary. The effects of selective breeding on life expectancy and age-specific mortality differed between females and males. In females, sedentary selection mice had early and high initial adult mortality and thus the lowest increases in mortality with age. Active selection females had the lowest early adult mortality, had limited mortality during midlife, and exhibited rapid increases in mortality rates at the very end of life; thus, they had deferred senescence. Median life expectancy was greater for both groups of selection females than for the two complementary groups of control females. Like females, sedentary selection males had the highest early adult mortality, and slow but steadily increasing mortality over the entire lifetime. Unlike the active selection females, active control males had the lowest mortality across the lifespan (until the end of life). Interestingly, the males with the lowest median life expectancy were those in the active selection treatment group. In both sexes, running (km/week) decreased over the lifetime to very low and virtually equivalent levels at the end of life in control and selection mice. Overall, these results demonstrate an evolutionary cost of selective breeding for males, regardless of exercise level, but a benefit for females when they have an outlet for the up-selected behavior. We conclude that correlated evolution of senescence occurs in mice selectively bred for high voluntary wheel running; exercise per se is beneficial for control mice of both sexes, but the impact on the effect of selection depends on sex; and the behavioral effect of exercise selection at an early age declines throughout the life span, which demonstrates decreasing genetic correlations over age for the genes involved in increased exercise.