The measurement method of the bidirectional reflectance distribution function (BRDF) for a rough surface was analyzed. A method to establish the BRDF model of the target surface material by artificial neural network is proposed. The sample BRDF curves variating with the scatter angle under several incident angles are given, and some of the curves were selected and sent into the artificial neural network. The network was trained with the Bayesian regularizing method. The mapping relation model of BRDF with incident and scatter angles was obtained. The scattering distribution curves under the incident angles, which participated in or did not participate in the training, were calculated with the network model. It was compared with the measured curves, and the result shows that the model is correct and practical.