Gaofen-4 (GF-4) satellite is the first geosynchronous high-resolution optical imaging satellite developed by China, and it has high temporal resolution and high spatial resolution. Aiming at the characteristics of GF-4 satellite data, we propose a cloud and cloud shadow detection algorithm combining spectral analysis and geometrical algorithms. Geometrically corrected and radiometrically calibrated GF-4 images are used to identify potential cloud pixels using spectral difference analysis techniques based on the spectral characteristics of clouds and typical land surfaces. The cloud probability is calculated according to the difference of spectral variability rate of clouds and cloudless features. The geometrical relationship between clouds and cloud shadows is combined with the sensor parameters to identify the projective regions of cloud shadows. Then the image-based dynamic thresholds are set in the projection regions based on the spectral characteristics of the shadows to detect cloud shadows. This algorithm can better identify thin clouds, and significantly improve the cloud shadow detection accuracy. The visual interpretation method is used to verify the detection accuracy. It finds that cloud pixels recognition in different regions are more accurate and the shapes are relatively complete. Compared with the method of cloud and cloud shadow matching, the dynamic-spectral-threshold algorithm proposed in this paper is more accurate in detecting cloud shadows.