The going concern concept is the basic prerequisite for company’s business activities and its forecasting is essential for the entire business community as well as for the large number of stakeholders participating in company’s value chain. The key success factor in going concern forecasting is the selection of the most appropriate model that can factually and accurately generate efficient going concern prediction, thus enabling the key decision makers to make the right decision. The going concern prediction is mostly connected to audit engagements, but in time of fast changes in global economic environment and corporate finance, going concern prediction has, in the last 35 years, become one of the global key economics research topics. The purpose of this work was to examine the effectiveness of the going concern forecasting models based only on quantitative financial indicators in the Croatian economy and to propose the new model. The model should be based on both quantitative and qualitative indicators, considering major Croatian economy business characteristics in a sustainable economy environment, especially after financial crisis in 2008. The main goals of the thesis were to explain the effectives of traditional and contemporary going concern prediction models, examine the effectiveness of Altman Z’-score model, Zmijewski model and Ohlson model and to propose the new going concern prediction model based on selected financial, quantitative (non-financial) and qualitative variables developed with the support of artificial neural network method. Empirical research and efficiency analysis performed on the selected three statistical models were conducted on the data sample from companies operating in Croatian economy. Conducted research and analysis confirmed the two supporting hypotheses of the work proving that traditional models are actually efficient in business performance prediction in Croatian economy, especially for the successful companies, but that there is a difference in the models’ accuracy. In general, there are several main challenges in the traditional models’ implementation in business practice. The first problem regarding traditional forecasting models users come across is their application in difference specific macroeconomic environments and national economies due to their general application set up. The second main issue is the models’ accuracy rate VIII in business failure or going concern forecasting due to the fact that the models are based on traditional statistical methods (MDA, probit and logit) that are not always adequate for forecasting complex and dynamic situations. The third problem is the fact that majority of traditional models is based solely on financial business indicators resulting from historical accounting data. All things considered, these models do not ensure appropriate sensitivity and cannot guarantee early warning signals to models’ users. The need for the new model, that would take into account all disadvantages of the current models, came out as the result of the empirical research. In the last several years, artificial neural networks method is being suggested, both in theory and practice, as more appropriate and efficient method than traditional statistical methods, without respecting all restrictive statistical prerequisites of traditional methods. Therefore, artificial neural networks statistical method was used in this empirical research in order to define new forecasting model, as well as logistic regression to define the most important variables in the forecasting model. The main results of the thesis is the new model and algorithm, based on the backpropagation algorithm of the artificial neural networks, for the company’s going concern forecasting. The most important variables for the forecasting are: total debt / equity, total debt / (equity + reserves), long-term liabilities / total assets, EBIT / interest, total assets / (equity + long-term liabilities). The new defined model demonstrated the accuracy rate of 97.69% in companies’ going concern prediction, which is higher than at traditional statistical methods analyzed in the thesis by which the main thesis hypothesis was confirmed: “Contemporary forecasting model, based on selected quantitative and qualitative variables by statistical method of neural networks and taking into account the specificity of the Croatian economic environment, can more efficiently assess companies’ going concern in comparison to traditional models, thus contributing to the stability and sustainability of the national economy.” Data used in the forecasting model definition was collected in the Croatian economy and Croatia, as the country, successfully represents transitional countries. This makes the new model applicable in regional transitional countries with similar macroeconomic and institutional conditions.