In this paper, we study the size and power of permutation test for location param ctcrs when the sample size is small and the population distributions are non-identical.It is compared with some popular test methods including t-test, Wilcoxon test, and bootstrap method.It is found that the permutation test based on the statistic of sample mean difference is not inferior to the others in terms of type-Ⅰ error and is more powerful than t-test and Wilcoxon test.A simple continuation correction method is proposed to regain the precision of size under discrete populatioas Exten sive Monte Carlo simulation is carried out to demonstrate our findings.