In this paper authors analyze the predictive power of existing models for the detection of earnings management of the companies in countries around the world. In addition, based on a representative sample of companies in the industrial sector of the Republic of Serbia, they tend to develop an improved model for the detection of manipulative financial reporting, which will be suitable for the Serbian business environment by using multiple linear regression. The application of existing models of discretionary accruals in the Serbian economic environment shows that these predictive models do not have sufficient explanatory power (Jones model 5.4%; Dechow model 2.6%; Kasznik model 37%), and there is a need for their further modifications. From these models, only Kasznik model proved to be statistically significant model for the sample of 65 companies in the industrial sector of the Republic of Serbia. As a result of the research, a modified model of discretionary accruals has been developed, which has improved explanatory power and which allows detection of new techniques of manipulation of financial reporting in the industrial sector of the Republic of Serbia. The explanatory power of our modified model is 63.7%.