Since there are few tail data when the generalized Pareto distribution is applied to the data beyond the threshold, the Bootstrap confidence interval for the parameters is derived by combining the maximum likelihood estimation of the parameters with the Bootstrap method. The Bootstrap method can not only make full use of the sample information by resampling, but also avoid the estimation of the variance of the estimation when there are fewer samples. Bootstrap confidence intervals for parameters are proved being asymptotically efficient. Numerical simulation results show that the interval length and coverage are reasonable under a certain confidence level. The established Pareto model is used to analyze the magnitude data of the Bayan Har seismic zone. The results show that under the same confidence level, the length of the confidence interval determined by the Bootstrap method is shorter than that calculated by the approximate covariance matrix, which supports the applicability of the generalized Pareto model to describe the earthquake magnitude.