Abstract: Cloud cover status determines whether satellite data can be used for retrieving or not. Cloud screening is essential in data processing flow of greenhouse gases retrieval. The threshold, whether appropriate or not, determines the accuracy of cloud detection in O₂-A band via threshold method, methodology of how to obtain optimal threshold becomes a key-point requirement. Thresholds are selected based on empirical statistics that hence lack of systematism in most cases. A semi-empirical algorithm of threshold selection has been proposed, the algorithm construct a function by combining clear-sky contain ratio and false determinant ratio, then selecting the optimal threshold through seeking the maximum of this function. Finally, the algorithm was validated by GOSAT observation data and the result shows that, comparing with traditional statistical method, threshold determined by this algorithm increasing the cloud detection accuracy near upon 10 percent when applying it to O₂-A band cloud detection.