This paper studies optimizing neurofuzzy system using a hybrid approach of a modified probabilistic incremental program evolution algorithm (MPIPE), random search algorithm, and evolutionary programming (EP). Neurofuzzy system is a combination of fuzzy system and neural network. The performance of a neurofuzzy system depends largely on selection of fuzzy membership functions, partition of input space and fuzzy rules. Two neurofuzzy models, additive and direct, are proposed in which neurofuzzy system calculation is based on tree structural representation. Without prior knowledge of the plant, parameters of fuzzy membership functions, the number of fuzzy rules and weights of neurofuzzy system are optimized using a hybrid method of MPIPE and EP algorithms simultaneously. Simulation results for identification of nonlinear systems show the feasibility and effectiveness of the proposed method.