Transit agencies are looking for analytical tools to improve their rolling stock maintenance programs and fleet size. If agencies underestimate the number of buses out of service for maintenance at any given time, they cannot adequately cover all scheduled service. If agencies overestimate the number of buses requiring maintenance, they overcapitalize the rolling stock fleet, resulting in an inefficient allocation of scarce funds. This article presents a methodology for using a fully parametric duration model to determine the expected amount of time a vehicle is in service and out of service. Additionally, the transition probabilities are calculated using the hazard data to determine the probability of transitioning from an in-service state to an out-of-service state at a given point in time. An empirical analysis is conducted using data from the San Francisco Municipal Transit Agency and duration models used to identify the transition probabilities from in-service to out-of-service states and the optimum fleet size.