An in-orbit calibration model has been adapted to improve the sun sensor's in-orbit accuracy. This model consists of 2 sub-models. One is the measurement model of the sun sensor, the other is pseudo measurement model. In the measurement model, variety of in-orbit factors have been taken into consideration, and the measurement model has been modified to compensate the errors caused by the in-orbit factors. While in the pseudo measurement model, the accurate sun ephemeris and the high-precision output of star sensor are combined together to calculate the true sun vector in spacecraft-body coordinate system. The accurate sun ephemeris is used to calculate the true sun vectors in GCRS (Geocentric Celestial Reference System), the attitude data from star sensor is used in coordinate system transformation of the sun vector. The pseudo measurement model has been analyzed considering varies errors, and the result shows that the errors would not have effect on the calibration model. The least squares regression algorithm is applied in this calibration model to optimize the internal and external parameters. The calibration model has been verified on a sun sensor which was installed on a satellite and launched in September 2015. The result shows, without in-orbit calibration, the accuracy of the sun sensor in-orbit decreased to 0.22 degrees in x direction and 0.27 degrees in y direction, while the accuracy was 0.11 degrees and 0.14 degrees before launching. With the in-orbit calibration method applied, the accuracy of the sun sensor would rise to 0.15 degrees and 0.18 degrees.