This thesis considers a queueing network model for a flexible manufacturing system with limited input and output buffers, where machine blocking is allowed and two AGVs are used for input and output material handling, respectively. The queueing network model is danalyzed approximately by decomposing the manufacturing system into two subsystems such as the first-level queueing system and the secondlevel queueing system. The first-level queueing system (representing a subsystem composed of input material handling AGV, and parallel input buffers and workstations) is further decomposed into separate queueing subsystems which can be analyzed in isolation. The second-level queueing system (representing a subsystem composed of output material handling AGV, and parallel output buffers) is treated as a queueing system but under revised queue capacity and arrival and service processes. As the system performance measure, both the system throughput and the queue length distributions are considered. For the marginal queue length distribution of the second-level queueing system, an efficient algorithm is especially exploited by characterizing the reversibility of the system rather than by trying any conventional way of solving balance equations. The algorithm is illustrated with numerical examples.