Two types of economic replacement policies for a fleet of vehicles are investigated. One is a simple group replacement policy in which vehicles are replaced at a pre-set age or milage. In the second, a repair limit is set that varies with the condition of the vehicle. A vehicle is replaced whenever it requires a repair expected to cost more than this allowable repair limit. The structure of the repair limit problem is shown to be a Markov decision process, and steady-state repair limits can be determined by R. A. Howard's policy improvement routine. Because for each state the repair limit is a continuous decision variable, however, Howard's routine is modified to allow a search procedure to determine the optimal limits. Using data for Postal Canada vehicles, these two types of policies are compared and the results indicate the superiority of repair limit policies over simple group replacement.