Many studies have been conducted on the empirical relationship between the positioning accuracy and the convergence and bisector elevation through reversely analyzing remote images. This paper proposes a positioning accuracy model based on the spatial forward intersection from the perspective of remote satellite designing. This work quantitatively calculates the positioning accuracy from stereo images with different convergence and bisector elevations， including the asymmetry and roll angles. Our study demonstrates that using this selection principle， the best multi-view stereo image can be identified， and the accurate ground coordinate for three dimensional modeling of cities can be achieved.