This paper presents a framework for obtaining better performance from multiple model approaches for estimating infrequent abrupt changes in nonlinear systems. By using insight into the nature of the problem and basic probability, a procedure that greatly reduces the number of filters required by multiple model approaches is obtained. This allows for a much longer detection horizon without increasing the computational requirements and results in improved performance over standard multiple model approaches. The performance of this approach is evaluated for state/parameter estimation of a heptane to toluene aromatization process. The method is also shown to be robust to errors in the assumed noise statistics.