The maintenance effect is a peculiar factor applied to repairable systems such as rolling stocks. Conventional statistical analysis for failure times takes into account one of the two following extreme assumptions, namely, the state of the system after maintenance is either as "good as new" (GAN, perfect maintenance model) or as "bad as old" (BAO, minimal maintenance model). Most of the papers concerning the stochastic behavior of railroad systems assume two types of maintenance: perfect and minimal maintenance. However, Lee, Kim & Lee (2008) analyzed the failure data of a door system in Metro EMU and the effect of preventive maintenance was imperfect. It is seen that the imperfect maintenance is of great significance in practice. This article describes how to deal with the maintenance effect in reliability studies of rolling stocks. Maintenance policies under imperfect maintenance are described and the method is proposed to evaluate their performance.