Given the nature of banking industry activities, which mainly relate to the allocation of resources, credit risks are increasingly faced by this industry. Therefore, knowing the origin of credit risk and its estimation are always fundamental issues for this industry. In this regard, in order to solve this problem in Bank Sepah, this study aims to identify the features effective to credit risk of real customers of Bank Sepah as well as to design a model to predict the probability of credit risk default by employing genetic algorithm and probit regression models. The data of this research is gathered using the credit facilities paid to persons in 2016. Among all the credit facilities paid to real people in 2016, two samples each of size 3600 (for fitting the model) and two sample each of size 400 (in order to verify the model by the ROC curve) were randomly selected. Meanwhile, MATLAB software has been used to analyze the data. The result of the study shows that the genetic algorithm (GA) method has the ability to determine the variables in three difference levels based on the degree of importance. The results also show that area under the ROC curve in the GA method is equal to 0.92 but in the probit regression method, it is equal to 0.72 demonstrating the higher ability to predict the likelihood of facility failure in the GA method compared to the probit regression method. The results of the ROC curve verification show that the GA correctly predicts 91.8% of cases compared to 90.0% in probit regression.