The number of times individuals diagnosed as schizophrenic are hospitalized varies greatly. Four theoretical models are proposed to account for the variation. The positive reinforcement model proposes that individuals are positively reinforced at each episode, and is consonant with theories of institutionalization. The negative reinforcement model proposes that individuals are negatively reinforced at each episode, and is in aggreement with the idea of therapeutic efficacy. The heterogeneity model proposes that individual differences account for the variation. The pure random model proposes that the differences are explained by a random process. A negative binomial distribution, consonant with both the positive reinforcement and the heterogeneity model, is fit to the distribution of numbers of hospital episodes of schizophrenia for a cohort from the Maryland Psychiatric Case Register. Correlations between occurrence of episodes in early years with occurrence of episodes in later years also suggests that the pure random and the negative reinforcement models are not operative. Further analysis suggests that the heterogeneity model is the correct one.