Long-time series stable DMSP/OLS nighttime light data lack comparability and include a lot of saturated pixels in the bright cores of urban areas. The two problems have limited applications of the dataset in spatial analysis and temporal comparison. To help address the two problems, this article developed a classified correction method for Stable Nighttime Light （SNL） imagery across China by using invariant regions for calibration. The SNL images are divided into saturated pixels and unsaturated pixels. For saturated pixels, five cities such as Beijing were selected as the invariant regions, and the radiance calibrated nighttime light images （RCNTL） without saturated pixels were selected as the reference. Based on the assumption that the correlation between unsaturated pixels of the invariant regions could also be applied to saturated pixels, the reference images were used to correct saturated pixels of the SNL images. Meanwhile, the saturation corrected saturated pixels were calibrated based on the interclibration relationship between the different RCNTL images. For unsaturated pixels, 13 cities including Changsha were selected as the invariant regions. According to the overall time change trend of the sum of the unsaturated pixel values, the 13 SNL images were determined as the references, and other SNL images were intercalibrated based on the correlation between unsaturated pixels in the invariant regions. To validate the accuracy of the calibration results, a variety of intercalibration and saturation correction methods were adopted for comparison. The classified correction method proposed in this paper was found better for the calibration of F10, F12 and F14 satellite imagery than the other two intercalibration methods. The other two methods had better calibration results for F15 and F16 satellite imagery. The three methods generally achieve intercalibration of the dataset imagery and make the intercalibrated images comparable. Compared with the Vegetation Adjusted NTL Urban Index （VANUI）, the saturation correction result of the classified correction method solves the problem that the saturated pixels concentrate in the central area of cities. The result reduces the saturation of the pixels and is a closer fit to the RCNTL images. Besides, the corrected SNL images have relatively good correlations with GDP and power consumption values, and can reflect regional economic development differences more objectively.