Taking Shiyang River Basin as an example, the models of GCAWI, spatial autocorrelation index, multiple(single) centre exponential with the feature of spatial autocorrelation are applied to achieve three goals: the conversion from township unit to grid unit, the regulation of determining appropriate grid size and spatial distribution patterns of population density. It turned out that:①The population density distribution is relatively scattered but concentrated in Shiyang River Basin,having a spatial structure pattern of point(three points)line(four lines) region(three regions);②The level of global spatial autocorrelation is improved by applying different grid scale size.The index of Moran’s I Shows larger difference and contingency;③The spatial distribution of population density has positive spatial autocorrelation in Shiyang River Basin.And the range   from 8 000 to 10 000 metres is the best choice to present the basin’s spatial distribution characteristics of population density;④The multiple and single centre exponential model with the feature of spatial autocorrelation has higher significance than traditional exponential model,but changes the properties and size of distance attenuation coefficient. The difference between the two kinds of models is caused by the population density centre of Jinchang.