The present economic and financial crisis has underlined the importance to financial institutions and investors of having access to efficient methods of quantifying credit risk, or the probability of default. The logit models are among the techniques commonly used by large organizations and rating agencies for predicting insolvency. However, it should be borne in mind that some problems arise when using these models, such as the selection of the explanatory variables or the composition of the sample from which the model is obtained. These aspects have a decisive influence on the prediction models used to quantify companies’ credit risk. The present study describes the problems that arise with logit on a sample of Spanish companies and shows that the estimated prediction models are indeed modified by changes in the sample on which they are based.