The current maintenance practice at most manufacturing plants is to conduct maintenance tasks during non-production shifts, breaks, or weekends, which may unnecessarily introduce extra labor and overhead costs. In order to reduce such costs and make more efficient use of maintenance resources, it is important to look for hidden maintenance opportunities to perform short-duration maintenance tasks, while not bringing any short-term production losses. In this paper, we establish analytical approaches to compute stochastic maintenance opportunity windows (MOWs) for the unreliable two-machine one-buffer system with both discrete time and continuous time Markov models. Instead of allowing buffers to be empty as previous MOW models were constructed, we focus on computing a lower bound in a buffer to reserve buffer space for unexpected random failures during recovery phases. Furthermore, general trends of these lower bounds and their corresponding stochastic MOW values have been investigated through numerical case studies with various system parameter change .