Commercial banks need to assess the credit risk of customers when issuing loans to decide whether to approve loan applications. Traditional credit risk assessment methods mainly rely on customers’ financial statements and credit history records, which are often not comprehensive and accurate enough. Therefore, commercial banks need to find a more effective credit risk assessment method. In this paper, taking the German credit data set containing 1000 borrowers’ information as an example, the credit risk of commercial banks is assessed by using the account status of borrowers and other relevant data through the establishment of BP neural network model. Compared with Logistic regression model, discriminant analysis and Naive Bayes, BP neural network model outperforms other models in accuracy, recall and precision. The conclusions can provide theoretical reference and practical guidance for bank credit, risk early warning and prevention of default risk.