Existing algorithms use a single chemical kinetic model over the entire domain of a reacting flow simulation. However, it is well known that the chemistry varies significantly in different regions of a flame. For combustion and other reacting flows with very complex chemistry, computing all possible reactions and species everywhere in the computational domain (the full chemistry approach) is very computationally demanding. Here we propose an adaptive chemistry approach that avoids computing species and reactions in regions where they are negligible. In this approach, a set of reduced chemical kinetic models is used in a simulation, each reduced model being used only under reaction conditions where it is faithful to the full model. A procedure to implement adaptive chemistry is presented, and compared with the conventional full chemistry approach. Adaptive chemistry fairly accurately reproduces the full chemistry solutions, while reducing the cost of the computation both in terms of CPU time and in terms of memory requirements. This was demonstrated by simulating a turbulent planar hydrogen/air shear layer flame and two axisymmetric laminar co-flowing partially premixed methane-air flames. The challenges that need to be overcome to make this method more broadly useful are summarized. (C) 2003 The Combustion Institute. All rights reserved.