OBJECTIVES: This study used discrete-time survival analysis to estimate the tenure of primary care physicians in Community Health Centers (CHCs), to identify the changing risk of leaving Community Health Center employment as time passes, and to identify factors associated with a physician's likelihood of remaining in a Community Health Center. Because of dramatic differences in physician career trajectories, much of the focus was on differences between physicians with and without National Health Service Corps obligations. METHODS: Beginning with an administrative dataset at the Bureau of Primary Health Care that listed primary care physicians for each Community Health Center, the completeness and accuracy of the information provided were verified and an analytic database of all physicians working in those centers during a 21-month measurement window from January 1, 1990 through September 30, 1992 was constructed. The data included start and end dates, percent full-time equivalent status, and certain demographic characteristics. In addition, several data elements describing the Community Health Center were merged onto each physician record. These included urban or rural location, expenditure level, productivity, and federal grade. Through the use of discrete-time survival analysis, it was possible to include in the analytic sample all 2,654 physicians who worked during the period, even those who started working before January 1, 1990 and those who were still working on September 30, 1992. Survivor functions were estimated showing the proportion of physicians remaining after each quarter of their tenure (ie, after the fourth quarter of work, after the 12th quarter of work, etc). In addition, hazard functions were estimated showing the risk that a physician who had worked through the end of one quarter would leave during the following quarter. Finally, multivariate analysis demonstrated the relation of certain physician and center characteristics to the likelihood of the physician's leaving the center during each quarter. RESULTS: The median tenure of primary care physicians in Community Health Centers was approximately 3 years regardless of whether or not the physician had a National Health Service Corps obligation. But the career trajectories for the two groups of physicians varied dramatically. Most physicians left on or about their anniversary date, probably because it coincided with the end of their contract, but the effect was much more pronounced for National Health Service Corps physicians than for non-National Health Service Corps physicians. By the end of 5 years, approximately 36% of physicians who started without an National Health Service Corps obligation were still working compared with only approximately 17% of those with an National Health Service Corps obligation. CONCLUSIONS: The study demonstrates the value of discrete-time survival analysis in addressing questions related to the tenure of primary care physicians in Community Health Centers, making it possible to use data from physicians whose Community Health Center careers began before or ended after a give measurement window. Second, the study measured primary care physician tenure, providing center directors with a yard-stick against which to compare their own center's performance. Finally, the data provided some help in trying to explain differences in the propensity to stay or leave employment in Community Health Centers.