Bankruptcy prediction is a topic of numerous scientific papers and research. Successful bankruptcy prediction is of great importance for every-day practice, especially considering the financial crisis and dynamic market environment. In this paper models for bankruptcy prediction have been designed for Croatia, Bosnia and Herzegovina, Serbia and Hungary based on accounting information from financial statements from 2012. and 2013. Data was taken from the „Amadeus“ database. Companies were paired for each country, based on their business activities and the size of the company, or size of the operating revenue and total assets. Sample for Croatia consisted of 676 companies, or 338 paired companies on the basis of which ratios have been computed and a binary logistic regression has been conducted with the classification success of 90,6%. Also, the method of neural networks was conducted using multilayer perceptron which gave an even better performance in classification of 91,45%. Sample for Bosnia and Herzegovina consisted of 11 paired companies, or 22 companies in total following the use of binary logistic regression and neural network which gave a success in classification of 100%. Sample for Serbia consisted of 68 companies, or 34 paired companies. Based on the most significant accounting ratios using binary logistic regression classification accuracy of 97% has been obtained, which is a better result compared to the one obtained using neural network which amounted to 94,3%. Sample model for Hungary consisted of 34 companies, or 12 paired companies and, based on accounting ratios using binary logistic regression and neural networks, a perfect classification of 100% was achieved. Successful bankruptcy prediction can be achieved using binary logistic regression and neural networks where the level of accuracy for all models is extremely high, which means that it is possible to create an efficient model based on accounting information.