When comparing the traditional financial risk measurements, Value at Risk(VaR) has its benefits for providing a single number that summarizes the overall market risk of the portfolio. Considering the fact that VaR measurement is standardized as a tool for international market risk measurement, back-testing the accuracy and performance of the VaR models plays a crucial role. In this sense, this paper proposes a way to validate the accuracy of the VaR models. Firstly, Bayes factor is used to assess statistical accuracy of the models. For the next step, loss function is applied to measure the differences between the realized and expected losses. Through the procedure, back-testing VaR models considering both frequency and magnitude of violations and comparing between the models can be achieved.