Abstract: Objective To introduce a meta-analysis method based on permutation test of linear models for small-sample meta-analysis. Methods A Fleiss93cont example was used to introduce a SAS macro（%METAPERM2） for covariate analysis developed by Kromrey of the Southern University of Florida in USA. This small-sample dataset did not satisfy the assumptions such as normality, independence and homogeneity of variance. Results The regression coefficient of generalized linear model was X₁ （age）=0.125, X₂ （area）=0.291. The results of the five regression weight test methods were: The traditional weighted least squares （WLS） β₁=0.000, β₂=0.338, Freedman Lane model β₁=0.228, β₂=0.180, Kennedy model β₁=0.472, β₂=0.557, Manly model β₁=0.064, β₂=0.040,Ter Braak model β₁=0.075, β₂=0.142. Conclusions Based on the hypothesis of normality, independence and homogeneity of variance, the significance of the traditional WLS coefficient test was larger than that of any permutation test, and permutation test may be a more suitable statistical method for small-sample meta-analysis.