Focusing on the problem of threshold uncertainty in tail-overbounding model, non-par hypothesis test method was adopted to evaluate the fitness of threshold which was the key to obtain a reliable estimation of GBAS protection level integrity risk. The P value calculated from the test was used to choose the best threshold and bootstrap method was taken to compute its confidence interval. Simulation based on a standard error model discussed the influence of various thresholds on integrity risk and the performance of different test methods were compared. Results also show that: compared with traditional tail-overbounding method, the model with threshold chosen by hypothesis test gives a more accurate description on tail samples and the corresponding integrity risk value proves to be more stable when threshold varies in its confidence interval.