Abstract: Objectives: For the satellite images with large attitude measurement error, wide field angle and weak intersection, the traditional block adjustment of rational function mode (RFM) has limited error compensation ability. Although establishing error model from the object-space can avoid the limitations, the model construction of existing methods is more complicated. Methods: In this paper, the line-of-sight vector of image point is recovered by RFM. Then, the error model for the line-of-sight direction angle is constructed, with a number of parameters consistent with the traditional method. Finally, the model parameters are solved by block adjustment. Results: Experiments on simulated and real GF-1 Wide Field View (WFV) satellite images show that the proposed method can always maintain an image-side positioning accuracy of about 1 pixel under different angular element errors. When using 13 control points, the plane positioning accuracy can reach 12.63 meters, and the elevation positioning accuracy can reach 17.20 meters. Experiments on GF-2 satellite images with intersection angle of only 0.05 degrees show that this method can obtain an adjustment convergence result with an image-side positioning accuracy of 1.89 pixels. Conclusions: Compared with the traditional methods, the method using line-of-sight direction angle correction can achieve better and more stable positioning results for the satellite images with large attitude measurement errors and wide field angles, and can obtain convergence results without weak intersection detection and digital elevation model assistance in the block adjustment under weak intersection conditions.