On the basis of the constructed evaluation index system of credit guarantee risk, the neural network technology is applied to comprehensive risk assessment of the credit guarantee products. The training samples，verification samples and testing samples that the model needed are generated by random sampling method in the range of single-index evaluation standard. The case study indicates that the methodology for generating samples and the process for establishing BP-ANN model are effective and reliable. The phenomenon of over-training and over-fitting can be effectively avoided, and the BP model possesses good generalization and is not influenced by human factors.