A new technique is presented for diagnosing faults in a Discrete Event System (DES) when the state of the system under diagnosis (SUD) is uncertain upon commencement of the diagnosis process. Specifically, a diagnoser is developed that detects, identifies, and isolates faults in a DES relative to an initial belief state: a set of states in which the SUD could be located at the diagnoser’s time of activation. The diagnoser does not need to be synchronously initialized with the SUD. Rather, it can be activated anytime before or after a fault in the SUD occurs, as it does not require knowledge of the SUD’s behavior pre-activation. This form of analysis can be understood as fault diagnosis under uncertain initial conditions. The construction procedure for the diagnoser is given and the new concept of semi-asynchronous diagnosability is introduced. The latter requires all failures in the SUD to be diagnosable by observing the behaviors of the SUD after the diagnoser is activated. Moreover, necessary and sufficient conditions for semi-asynchronous diagnosability of a DES are provided. The semi-asynchronous diagnosis technique presented herein is compared with classical synchronous diagnosis approaches. Illustrative examples are provided in order to explain the introduced concepts and the developed approach.