Purpose: This study determines the relationship between a medical school applicant's rural background and the likelihood of rural practice using different definitions of rural background. Methods: Cohort study of 599 physicians who entered the University of Virginia School of Medicine in 1990-1995 and graduated in 1994-1999. The "rurality" of the applicants' backgrounds was assessed by coding applicants' high school, college, and permanent addresses using 4 definitions of rural. In addition, most physicians responded to a survey question "Did you grow up in a rural area?" The same 4 definitions of rural were used to assess the rurality of practice locations. Logistic regression models were used to predict the simultaneous effect of different definitions of rural background, gender, and applicants' career preference at matriculation on the probability of practicing in rural areas. Findings: In univariate analyses, the high school, college, and permanent addresses were all predictive of rural practice using 1 or more definitions of rural. In the multivariate analysis, only the "grew up rural" self-description was predictive of rural practice location using 3 of the definitions of rural. In a secondary analysis, the grew up rural self-description and applicants' career preferences at matriculation were predictors of rural practice. Conclusions: Readily available addresses were predictive of rural practice. In a multivariate analysis, physicians' self-description about having grown up in a rural area was the best predictor of rural practice. Recruiting more applicants who match this definition of rural background should increase the number of rural physicians.