The lumped scheme with consideration of catalyst deactivation was adopted to simulate catalytic cracking of gasoline for predicting the product distribution of secondary reactions. A reactant oriented selective deactivation model was developed using a new strategy of selective deactivation coupled with nonselective independence deactivation. Catalyst deactivation was correlated with the time on stream rather than coke content. Furthermore, a hybrid self-adaptive genetic algorithm (termed AGA/SA), which incorporated evolution strategies and simulated annealing into a genetic algorithm, was developed and applied in parameter estimation of the proposed model. Results suggest that the lumped kinetic scheme incorporating a deactivation function could enhance its fundamentality and the predication accuracy. AGA/SA exhibits the desired improvements such as rapid convergence, high efficiency, strong in hill-climbing, and effective in escaping the local optimum. Good agreement between the predicted results and experimental data indicates that the proposed kinetic model for secondary reactions of fluid catalytic cracking (FCC) gasoline is well established and AGA/SA is reliable.