In a distributed system, process synchronization is an important agenda. One of the major duties for process synchronization is mutual exclusion. This paper presents a new centralized fault tolerant distributed mutual exclusion algorithm based on Agrawala and El-Abbadi's algorithm. In new algorithm, once coordinator crashes, algorithm can recover lost data and return the coordinator in earlier situation. Thus fault tolerance will ascend and centralize algorithm's "single point of failure" will be omitted. So based algorithm will be more reliable. The only trade off is consuming some inappreciable time in case of coordinator's crash.