The combination of measured data, rock physics and multivariate statistical analysis in multidimensional space is helpful for interpreting the data quantitatively and reduces the prediction risk of reservoir framework and fluid information. Densities of quartz, clay and fluid were inversed with the density, clay content and measured porosity pressure of typical water-saturated sandstone samples (Dehua Han’s G group sandstone samples) in this paper. Compared to traditional binary regression method in two dimensional space, this method could not only avoid the effect of singular point in regression analysis, but also get fitting density which has more significant correlation with measured density.