Optimization of reaction paths for enzymatic systems is a challenging problem because such systems have a very large number of degrees of freedom and many of these degrees are flexible. To meet this challenge, an efficient, robust and general approach is presented based on the well-known nudged elastic band reaction path optimization method with the following extensions: (1) soft spectator degrees of freedom are excluded from path definitions by using only inter-atomic distances corresponding to forming/breaking bonds in a reaction; (2) a general transformation of the distances is defined to treat multistep reactions without knowing the partitioning of steps in advance; (3) a multistage strategy, in which path optimizations are carried out for reference systems with gradually decreasing rigidity, is developed to maximize the opportunity of obtaining continuously changing environments along the path. We demonstrate the applicability of the approach using the acylation reaction of type A beta-lactamase as an example. The reaction mechanism investigated involves four elementary reaction steps, eight forming/breaking bonds. We obtained a continuous minimum energy path without any assumption on reaction coordinates, or on the possible sequence or the concertedness of chemical events. We expect our approach to have general applicability in the modeling of enzymatic reactions with quantum mechanical/molecular mechanical models. (C) 2004 American Institute of Physics.