This paper develops a model to study the stay/leave decisions of Air Force officers. The model includes the most important institutional factors affecting an officer's career: promotion probabilities and timing, regular force integration probabilities, and mandatory separation and retirement probabilities. The model considers the officer's income for each potential combination of future grade and year of service and his civilian income opportunities. The model studies the incentives to retire for the current nondisability retirement system and the recent proposal by the President's Commission on Military Compensation. Numerical results are presented using actual data from Fiscal Year 1970 for nonrated officers who entered the Air Force through ROTC. On average, the officers' stay/leave behavior conforms with the model's predictions. Dynamic programming is the methodology used to construct the stay/leave model. Two versions of the dynamic programming model are considered. First, a sequential decision model for a risk-neutral agent is developed. The second version of the dynamic programming model generalizes the risk-neutral case to encompass risk-aversion. A functional equation is developed in which there is temporal resolution of uncertainty with respect to both military and civilian incomes.