Prediction of business failure is always a topical issue and has been carrying out many researches for many years using various statistical methods in order to shape the model with the most accuracy in predicting a business failure. This thesis explains the importance of prudent information, main financial statements, financial indicators and presents research carried out for the purpose of creating a basic model of business failure prediction and comparison with separate models for predicting business failure for different size of companies using the amount of total assets as criteria for sorting the companies in groups. The research was conducted based on available financial information of 989 companies in the manufacturing sector in Croatia. The financial indicators used as variables in the prediction models are: ROA, ROE, EBIT margin, asset turnover, current liquidity and self-financing rate. The empirical research has shown that the accuracy of predicting a business failure can be increased by sorting companies into different groups by the amount of total assets and by applying the logistic regression method to each group separately. The accuracy of predicting a business failure in models adjusted for companies of certain asset amount was higher by 3.6 percentage points to 18.10 percentage points for particular groups compared to the basic model used for all sizes of companies.