Abstract: We selected the grid of 5 m × 5 m in a dynamic monitoring plot (340 m × 300 m) as the sampling unites and chose 5 structural characteristics (density, average crown breadth, coverage, conspicuousness and average height) to study the spatial heterogeneity of community structure of Picea crassifolia forest in Dayekou Basin of Qilian Mountains by the fractal geometry and geostatistics methods. The results showed that the order of spatial variation in these characteristics was: density > average crown breadth > conspicuousness > coverage > average height, with the variation coefficient ranging from 43.7% to 79.6%. Moran’s  I index indicated that the structural variables had different degrees of spatial autocorrelation, and the order of autocorrelation was density > average height> coverage > average crown breadth > conspicuousness, with the range of -0.047-0.382. The exponential semivariation model well fitted the spatial variability in different structural features, and the range was 24.6-68.1 m. The variables displayed moderate spatial autocorrelation except for coverage, while the other variables had strong spatial autocorrelation, and the fractal dimension of the variables was close to 2, indicating a low spatial dependence among variables. The variables presented a superposing characteristic of zonal and patchy structures except for density and coverage, while the other variables presented strong patchiness property. Density and coverage had a certain spatial dependence on average crown breadth, conspicuousness and average height. Density and coverage for the spatial heterogeneity of community structural of P. crassifolia forests were 10 m and 0.5 hm², respectively.