The difference in statistical scales will bring different statistical results, so it is important and necessary to select statistical grids for the geographical conditions information statistics. This paper proposes a method of selecting statistical scale of geographical information with spatial autocorrelation being taken into account. By using geographic conditions census data and taking building information statistics as an example, this study gets building statistical information at different scales from 100m to 10000m, and meanwhile analyzes change trends of spatial autocorrelation of building information at different scales to make statistical scale selection. The results show that for building information statistics, 1000m is the turning point of statistical scales, and can be used as a suitable statistical scale in city region.