Influenced by external heat flow changes in space environment, on-orbit camera on satellite generates the geometric distortion, which causes positioning errors of remote sensing images. Classic geometric analysis methods for on-orbit satellite not only require the reference images and DEM models, which are need to be higher spatial resolution and geometric precision than the under-detection images, but also cannot separate camera distortion from detected geometric errors. In this paper, we purposed a geometric distortion detection approach for on-orbit cameras. Based on HJ-2A satellite’s 16 m CCD cameras, firstly, we chose the one-year images for the same area, North China Plain, which can remove the elevation error effect. Secondly, we chose one-month images as the reference images, and use SIFT algorithm to obtain the external geometric errors. Because of the same sensor for the under-detection images and reference images, the geometric errors caused by satellite platform and matching algorithm can be ignored. Therefore, the external geometric error is equal to the external distortion of on-orbit cameras. Thirdly, based on the external geometric errors, the least squares algorithm and statistics methods are used to obtain the internal geometric errors, which reveals optical distortion of the cameras. Experiment based on the real data of HJ-2A satellite’s four CCD cameras shows that the external and internal geometric distortion are affected by the external heat flow changes in different seasons, which provide basic data for the following camera distortion correction methods.