Using national databases of the Association of American Medical College, the authors employed logistic regression analysis to show the relative predictive influences of selected demographic, structural, attitudinal, and educational variables on the specialty careers choices of 1995 U.S. medical school graduates. Plans to pursue certification in family practice or an unspecified generalist career could be predicted with moderate success, while choices of general internal medicine and general pediatrics could not. The intentions of the 1995 graduates to pursue generalist specialty, were significantly associated with demographic factors such as female gender, older student age, and rural hometown; early interest in the generalist specialties; attitudes favoring helping people over seeking opportunities for leadership, intellectual challenge, or research; the presence of a department of family medicine in the medical school; and ambulatory care experiences in the third and fourth years. In the multiple-regression models used in this study, a number of factors widely touted as important to the cultivation of generalism were not significant predictors of generalist decisions; an institutional mission statement expressly addressing the cultivation of generalist careers; giving admission preferences to applicants who vowed an interest in generalism; public (versus private) school sponsorship; discrete organization units for general internal medical or general pediatrics; the proportion of institutional faculty in the general specialty of medicine and pediatrics; the level of educational debt; the students; clinical experiences in the first and second years of medical school. The authors acknowledge the danger of inferring causal relationships from analyses of this kind, and described how the power of previous associations--e.g., that between a required third-year clerkship in family medicine and graduates' family practice career choices--may be weakened when the independent variable spreads across institutional cultures that at present are less conductive to primary care. The findings of this analysis add to the evidence that generalist career intentions are largely carried on the tide of students' interests and experiences in family medicine and ambulatory primary care. In terms of the predictive values of the input variable in this study, career decisions for the other two generalist specialties--general internal medicine and general pediatrics--were essentially a crapshoot, either because the tactics to promote interest in these fields were ineffective (or confounded), or because the efforts were underdeveloped. Moreover, the statistical models of this study employed quantifiable variables that can be discerned and manipulated to guide the result, whereas medical students tend to identify less tangible elements as more powerful factors influencing their career choices. The results sharpen the strategic focus, but must be combined with those of other, descriptive analysis for a more complete understanding of graduating students' career decisions.