Multi-scale autocorrelation is an important basis of the study on spatial heterogeneity and spatial co-occurrence patterns of soil animals. A sampling field, 50 m × 50 m, was set up in farmland of the Sanjiang Plain Marsh and Wetland Ecological Experiment Station. The farmland had a history of more than 30 years of cultivation. The sampling field was divided into 100 subsamples and intensive samplings were done at the nodes of a 10 × 10 regular grid with 5 m inter-sample distances. Overall Moran’s I and Local Moran’s I were used to determine multi-scale spatial autocorrelations for mean density of the aboveground, belowground and above-belowground mite communities. Results show that the number of species and mean density of mite communities varied sharply between aboveground and belowground ones. Overall spatial autocorrelation analysis revealed that the mean density of the belowground mite communities showed significantly spatial autocorrelations was 35~40 m in August and 5~30 m in October, and of the above-belowground mite communities was 5~10 m, 35~45 m and 5~35 m, respectively, showing obvious multi-scale spatial autocorrelations between them and obvious multi-scale spatial autocorrelations also existed between dominant mite groups that were often large in number and commonly observed miter groups in the soil mite communities. Local spatial autocorrelation analysis indicates that significant local autocorrelations existed between the whole soil mite community and mite communities above the ground, under the ground and above-belowground in mean density, thus forming a horizontal distribution pattern with coexisting “patches” and “gaps”. All the above listed findings demonstrate that multi-scale spatial autocorrelation is an important feature of the belowground and above-belowground mite communities in farmland of the Sanjiang Plain, with mean densities of the mide communities and species forming a horizontal pattern with coexisting “patches” and “gaps”.