The objective of this paper is to compare the models that can be used as a tool in the bank examination and supervision process for the detection of banks, which are experiencing serious problems. Sample data of the study consist of 70 Turkish banks (38 banks failed during the period 2000-2008) and contain their financial ratios including capital adequacy, assets quality, liquidity, profitability and income-expense structure. By applying variable selection methods to the financial data, the important financial ratios are determined and these financial ratios are used as independent variables to obtain logit, probit and discriminant models. Finally, these models are compared to select the best model that has the highest predictive ability to differentiate between sound and troubled banks. Based on the results of the comparative analysis, the discriminant model is the best in classifying banks into non-failed and failed categories since it has the highest prediction success.