Abstract: This paper proposes a behavioral modeling of power amplifiers (PA) based on Improved Back-Propagation Neural Network (IBPNN) considering memory effects. In this model, Levenberg-Marquardt (LM) learning algorithm and training algorithm with momentum factors are adopted to update the weights and biases of BPNN. Comparing to the typical BPNN model, the proposed model can get higher accuracy with less training. Also, the time- and frequency-domain simulations of three-carrier WCDMA signal with 20MHz bandwidth using in this model exhibits good performance of the model. Moreover, the dynamic AM/AM and AM/PM characteristics obtained using the proposed model has demonstrated that the improved BPNN can track and describe the memory effects of the PAs well. Finally, an experiment with 16QAM OFDM signal and 20MHz bandwidth shows the universal of the proposed model.