Both Value-at-Risk and stress testing loss are important measures for the market risk. This paper develops a mixture of Generalized Extreme Value distributions and applies it to estimate these two measures coherently. Since this model can capture tails natures of the return distribution in details, the estimated risk measures are more precise than those of the Generalized Extreme Value distribution are. To assess the performance of this approach, this paper uses the Taiwan stocks market data in empirical study. The results and their implications in management are also discussed.