The results of financial condition analysis are used in the research on bankruptcy prediction of companies. The assessment of financial data quality involves also the detection of outliers. In the literature on bankruptcy prediction one can find deliberations on the problem of outliers. The proposals for solving this problem range from not taking any actions, through replacing or removing the outliers, to applying robust methods. Therefore, in the empirical research, some doubts concerning the choice of an appropriate approach to the outliers appear. The aim of the article is to present the outcomes of empirical research on the usefulness of selected techniques for identifying outliers in bankruptcy forecasting. In the study, both one-dimensional (based on centiles and Tukey’s criterion) and multi-dimensional (projection depth function) procedures of detecting outliers were considered. So as to assess the classification accuracy of chosen bankruptcy prediction methods for a test set, the sensitivity, specificity, accuracy and AUC measure were used. The analysis was based on data concerning manufacturing companies in Poland.