Abstract: Main reason of the existence of systematic error of POS data is analyzed. Error equation of POS-based relative orientation is deduced. Regression model to compensate the systematic error of POS is established. Three sets of aerial images in different scales with POS data are used for experiment. Experimental results show that the systematic error of angular elements in POS data is the main error source which causes large vertical parallax. Compared with original POS data, the POS data corrected by the regression model can significantly decrease vertical parallax. Accuracy of geo-referencing is considerably improved without block adjustment. Satisfying result of experiment verifies the correctness and feasibility of the regression model.