In many engineering applications, some phased-mission systems (PMS) may contain a large number of phases and repairable components. Traditional binary decision diagram (BDD) based methods or state-enumeration methods can suffer from the BDD explosion or the state explosion for this kind of PMS. This paper presents a non-simulation method for the reliability analysis of large PMS. In our approach, the system reliability is approximated by the system availability at discrete time. The discrete-time availability is modeled by the sampling of success states, which avoids the BDD explosion as the number of phases increases. Furthermore, BDDs are used to simplify success states, and enable our model to avoid the state-explosion problem. Two real-world PMS are analyzed to illustrate that the time and the storage cost of our approach do not increase exponentially with the number of components and phases in the PMS.