: We are concerned with a real-time quality of service (QOS) estimation method for ATM statistical multiplexers with heterogeneous ON/OFF sources. We propose a real-time modeling method for the superposed traffic by a two-state MMPP by approximating the density function of aggregate cell arrival rate as a Bellows-like Poisson process. We also investigate a real-time calculation of the waiting time distribution for MMPP(2)/D/1 system to consider an upper bound of the cell loss ratio (CLR) for the finite-buffer system. Numerical examples show that the proposed QOS estimation method gives a good estimation of CLR.