The empirical rock failure criteria play an important role in predicting the mechanical behavior of rocks, since affecting parameters on the behavioral properties of rocks are complex and numerous. Usually the empirical failure criteria are obtained by statistical methods, namely by means of fitting the mathematical function onto the experimental data. For evaluation and verification of the results, however, the quality and accuracy of the fit must be considered. This paper studies the efficiency and accuracy of conventional empirical rock failure criteria using a new approach, which is based on the analysis of uncertainty amplitude on the constant parameters of each criterion. The approach adopted here, is based on the combination of some parameters such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), type of function, degree of freedom and the confidence level of constants. Correlation coefficient (R2) is then a comprehensive test and can be used effectively for comparison and accuracy of empirical rock failure criteria.