In this study a genetic algorithm (GA) approach for determining new reaction rate parameters (A, beta, and E-a in the non-Arrhenius expressions) for the combustion of a hydrogen/air mixture in a perfectly stirred reactor (PSR) is assessed. A new floating-point coded GA and fitness function have been developed that dramatically increase both the rate of convergence and the predictive accuracy of the algorithm, thus promising the extension of the method to more detailed reaction schemes. Output profiles of species for 20 sets of PSR conditions, obtained from an original set of rate constants, are reproduced following a GA optimization inversion process. The new sets of rate constants following each iteration are constrained to lie between predefined boundaries that represent the uncertainty associated with the experimental findings listed in the National Institute of Standards and Technology (NIST) database. Comparisons with previous optimization work have demonstrated that those mechanisms generated using the NIST constraints can be applied to combustion scenarios outside those used in the mechanism's construction. In addition, the flexibility of the GA has been demonstrated by its success in generating reaction rate coefficients that reproduce a set of randomly perturbed species profiles.