The article contains an evaluation of the quality and universal character of three models of prognosticating insolvency, constructed on the basis of discrimination analysis and artificial neural networks. The evaluation of the created models was conducted on the basis of four companies operating in different lines of business, which declared bankruptcy in the years 2002-2003. The objective of the verification was to find out, how many years in advance it was possible to predict insolvency with the help of the models in question, and whether the line of business in which a given company operated might influence the efficiency of the said models.