For detecting the applicability of classical urban population density models on the microcosmiclevel, this paper calculates grid-based population landscape density of Shenyang city, analyzes it's spiatial auto-correlation and variability using theories and methods of geostatiatics. It is found out that all semivariogram functions of populatin density fit with spherical model with negget in ten kinds of grains from 100m to 1000m, indicating that population distribution presents structural characteristics in the spatial extension of Shenyang city. But because of different grains, the auto-correlation scales, neggets, and stills, the rate of negget and still is different. This means that spatial auto-correlation of population distribution depended on the scale intensively, resulting in scale effects.By calculating semivariogram functions curves of different directions, it can be found that population density takes on a classical zonal anisotropy, which means that there were different structural characteristics in different directions for poplation distribution. The analysis of population density auto-crrelation and variability should be taken as bases for researching urban population density distribution. The impacts of grain on population density auto-correlation and anisotropy are objective, so more attention should be paid to choosing feasible scale and identifying structure of population distribution. In the specific time for a certain city, the condition of population distribution is objective, while population density models are abstract expressions of population distribution characteristics, whose parameters are only quantitative indexes of expressing population distribution pattern. So, when researching spatial structure of population distribution in a city, the emphases should be put on how to identify spatial structural characteristics of population distribution, rather than on how to apply classical population density models mechanically.