The authors assess the importance of educational debt in graduates' primary care specialty choices, and the variety of mechanisms through which debt may influence career decisions. Logistic regression models were used to identify significant predictors of the primary care specialty choices made by the 1991 and 1992 graduates of U.S. medical schools. These predictors were debt itself; other financial indicators; certain medical school characteristics; certain practice location plans; certain demographic factors; aspects of academic performance; and students' predisposition to a primary care specialty. Data for this study were gathered from a variety of sources at the Association of American Medical Colleges and from the Health Education Assistance Loans program. Both direct and indirect effects of debt were identified under specific conditions. The study revealed complex relationships between debt and the other predictors identified. For example, debt operated in relation to the levels of the graduates' expected incomes; debt from subsidized loan sources was significant for women who chose general internal medicine; debt was important in choices of family practice; and debt by itself was significant for those planning to practice in the West and who chose general internal medicine. Also, seemingly opposing effects of debt occurred. For example, in the family practice model used in this study, the threshold effect of debt was positive, while the linear effect of debt above the threshold was negative. Such vriations help explain the conflicting findings of some past research. These and other findings prompt the authors to state that when investigating the effects of debt, it is not fruitful to ask what the effect of the debt is on all three primary care fields as a group. It is more appropriate to ask several questions, such as: under what conditions does debt influence specialty plans? Among which groups of students does debt have an impact on specialty plans? Are all of the primary care specialties similarly affected by the issues surrounding debt? Does the effect of debt change over time? The authors conclude by indicating possible policy implications of their findings.