Due to complex dynamics, it has been extremely difficult to model high power devices. A predictive model is constructed by using a backpropagation neural network (BPNN). The BPNN was applied to predict electrical characteristics of a reduced surface field p-channel lateral double-diffused MOSFET. Drain-source currents for applied drain-source voltages were measured with a HP4156A. Prediction performance of BPNN model was optimized with variations in training factors. With respect to the reference models, the optimized models demonstrated considerably improved predictions. Model predictions were highly consistent with actual measurements. Further improvement was obtained by constructing a modular network comprising multiple BPNNs. (C) 2004 Elsevier Ltd. All rights reserved.