This study describes a credit risk assessment system using neural network models based on back-propagation learning algorithm, in which the neural network training stage is performed with different data. Three neural networks were implemented and trained to decide whether or not to grant a loan. The results of this research show that neural networks are a suitable model for credit risk assessment that can perform efficiently in this field. In this article, it has been studied by using the analytical-descriptive method and by using the observation tool of people's files. For this purpose, a clustered random sample including 5319 customers who took loans from Refah Bank in the period of 2017-2018 was used. This dissertation has identified the factors affecting the risk of customer default by using the conventional models of survival analysis, including the non-parametric Kaplan-Meier model and the quasi-parametric Cox model. has been Including; A- Examining the relationship between the credit risk of loans received by customers individually (which corresponds to the individual analysis of securities risk) with the overdue claims of Refah Bank of Tehran province and B- Examining the relationship between the credit risk of the loan portfolio on a non-individual basis (which is equivalent to the portfolio risk analysis) securities) with outstanding claims of Refah Bank of Tehran province. The results of survival analysis techniques showed that variables such as loan amount, number of installments, number of children, education, age, type of job and job title on the curves of survival function and function affect the hazard rate. In short-term time horizons (for example, one year), the economic conditions of the society play a key role in the occurrence of defaults of this category of customers.