This paper investigates irregularities in financial statements by applying the Beneish and Roxas models to Polish firms listed on the Warsaw Stock Exchange from 2015 to 2020. The total sample included 110 observations. The sample comprised companies that had received an adverse or disclaimer opinion by the auditors, but had not been fined by the Polish Financial Supervision Authority (KNF Board). The control firms were selected based on the industry as selected by the standard industrial classification code and on the financial year, with minimizing the difference in the size of total assets. The results indicate that the Roxas model revealed greater accuracy than the Beneish model on the tested sample. The use of logistic regression allowed a modification of the Beneish model to align it with the conditions of the Polish market. The modified Beneish model showed greater accuracy for the tested sample and companies fined by the KNF Board.