Because of cloud or smoke in remote sensing images, the information of earth objects could not been read. The models of attenuation light and airlight had been analyzed. The advanced Retinex algorithm has been applied to remove cloud or smoke from the remote sensing images. The result shows that the advanced single-scale Retinex algorithm with small standard deviation of Guassian function in convolution kernel (SSRS) was effective for removing cloud or smoke. The cloud or smoke was in brighter region in the remote sensing images. In particular, the advanced Retinex algorithm SSRS was more effective for removing the thick cloud or smoke. The advanced Retinex algorithm SSRS had three steps. First, the complementary image of original image could be got and the brighter region could be reversed into darker region. Second, the advanced Retinex algorithm SSRS had been applied to enhance the darker region. Third, the complementary image could be got and the information of brighter region in original images had been enhanced.