The paper discusses the issue of forecasting bankruptcy of enterprises in Poland. Whether a given company will have the capacity to meet its financial liabilities, i.e. whether it survives on the market, is a matter of interest of many market entities, in particular suppliers, lenders, and owners. Due to the work-consuming nature of a full analysis of the financial standing of an enterprise, attempts have been made to develop methods that would lead to an instant and reliable diagnosis related to the financial standing of an organization, based on the smallest possible number of parameters. This kind of need underpinned the development of bankruptcy forecasting models.

In the paper, two methods of forecasting the threats of enterprise bankruptcy have been compared: the novel method utilizing artificial neural networks and the traditional discriminative analysis. The author of the research used data on 180 Polish manufacturing enterprises. The population has been divided into the teaching and testing samples. In both samples the ratio of bankrupt companies to the non-bankrupt ones is 1:1. Each of the 180 analysed enterprises has been described by means of 27 financial indicators and additionally by one non-economic variable - the geographic region of the company's operations. In the development of own artificial neural network models, the author used the following research approaches:

• Approach I-K1, where the set includes all 28 diagnostic variables,
• Approach II-K2, where all model input data have been determined on the basis of an analysis of the array of correlation ratios of individual diagnostic variables.

For the purposes of verification whether the artificial neural network model is more effective in the Polish economy than the discriminative analysis model, the author has compared the effectiveness of two discriminative analysis models by B. Prusak with that of K1 and K2 models from the testing sample. The models developed by B. Prusak (P1 and P2) have been derived from the same population of enterprises. Utilization of the same statistical data has enabled the author to conduct a reliable comparative analysis of the ANN models with the discriminative analysis models.