This article examines bankruptcy prediction models for non-public Polish enterprises, using data from 2017–2021 for 416 companies across trade, production, and services, sourced from the Emerging Markets Information Service (EMIS). Nine discriminant functions were constructed, each bespoke to industry nuances, applying a group mean equality test for financial indicator selection and a stepwise forward method for model building. Model efficiencies were assessed against those created from unprocessed and winsorized data, utilizing two winsorization methods to minimize outlier effects. The primary aim was to develop a model with the highest classification efficiency on a test set and compare it with models by Polish researchers. Results underscore the necessity of tailoring models to the specificities of the Polish market and industries, revealing differences in classification effectiveness. Despite employing a group mean test and stepwise forward analysis, the production sector model displayed lesser discriminative efficiency compared to Pociecha D1 and Mączyńska-Zawadzki G models. The study emphasizes the critical difference in the approach to variable selection and analysis, stressing the importance of a comprehensive approach to variable selection and the need to test models under diverse conditions to enhance their bankruptcy prediction capability.