In this paper, a new indirect georeferencing method for linear array imagery is presented. The new method fully utilizes the original orientation elements directly measured by geopositioning system carried on the satellite. First the systematic errors contained among linear and angular elements were modeled by second order polynomial functions depending on time. The constant terms compensate the shifts and angular drifts between the image system and the GPS and star sensor system. The linear and quadratic terms model the additional systematic errors. There are 18 parameters which should be
estimated for image georeferencing. Considering the strong correlation among those orientation parameters, they were
treated as virtual observations and the weights were assigned according to prior-knowledge such as the precision of sensor position and attitude observations. By this step, the value change of orientation parameters in adjustment process can be constrained reasonably. The new method has been tested on two scenes of CBERS-2-3 satellite images. In the tests, the location accuracies of 15~25m for planimetry have been obtained using the adjusted orientation elements,
greatly improved compared to the direct georeferencing results. Actually, the results should be even better if the measuring errors of image point can be further reduced.