A new method for in-service cell loss ratio (CLR) QoS estimation for asynchronous transfer mode (ATM) networks has been developed. For a typical CLR, a large number of cells must be observed before statistically meaningful results can be achieved. These results may be obsolete resulting in ineffective network management reactions. For a variety of arrival processes, many analyses have shown there exists a relationship between the logarithm of the cell loss probability and buffer size. For models that do not possess long-range dependence, this relationship is often linear. On the other hand, for the fractional Brownian motion model that captures long-range dependent traffic behavior, this relationship has a polynomial form. The proposed method uses these relationships and observations of cell loss for several small pseudo-buffers to characterize the behavior of the actual system. Specifically, a real-time technique to dynamically detect the failure of meeting a cell loss quality of service (QoS) objective has been developed. The method requires a short observation period and is suitable for in-service monitoring of CLR QoS. Simulation studies show the effectiveness of this method for both modeled traffic and measured network trace data.