Crime rates respond to age shifts in the population and to changes in current social conditions. A flexible generalization of the characteristic solution to the age-period-cohort problem shows that they also respond to differences in birth cohorts. I used age-arrest data for a representative panel of 16 states to estimate annual differences in cohort effects, then regressed them on 21 covariates spanning a wide range of social, economic, and environmental conditions. Cohort effects explained both the 1960–1990 crime increase and the post-1990 crime drop about as effectively as current conditions; the covariates explained about half of these effects. Cohort criminality was primarily affected by three conditions: relative cohort size; the prevalence of crime during childhood; and the capacity of families and neighborhoods to socialize children. Because all of these characteristics are trending in favorable directions, cohort effects – and crime rates – will probably continue to decline.