As for the satellite remote sensing data obtained by the visible and infrared bands inversion, the clouds coverage in the sky over the ocean often results in missing data of inversion products on a large scale, and thin clouds di?cult to be detected would cause the data of the inversion products to be abnormal. Alvera et al.(2005) proposed a method for the reconstruction of missing data based on an Empirical Orthogonal Functions (EOF) decomposition, but his method couldn’t process these images presenting extreme cloud coverage(more than 95%), and required a long time for recon- struction. Besides, the abnormal data in the images had a great effect on the reconstruction result. Therefore, this paper tries to improve the study result. It has reconstructed missing data sets by twice applying EOF decomposition method. Firstly, the abnormity time has been detected by an- alyzing the temporal modes of EOF decomposition, and the abnormal data have been eliminated. Secondly, the data sets, excluding the abnormal data, are analyzed by using EOF decomposition, and then the temporal modes undergo a filtering process so as to enhance the ability of reconstruct- ing the images which are of no or just a little data, by using EOF. At last, this method has been applied to a large data set, i.e. 43 Sea Surface Temperature (SST) satellite images of the Changjiang River (Yangtze River) estuary and its adjacent areas, and the total reconstruction root mean square error (RMSE) is 0.82°C. And it has been proved that this improved EOF reconstruction method is robust for reconstructing satellite missing data and unreliable data. As for the satellite remote sensing data obtained by the visible and infrared bands inversion, the clouds coverage in the sky over the ocean often results in missing data of inversion products on a large scale, and thin clouds di? Cult to be detected would cause the the Alvera et al. (2005) proposed a method for the reconstruction of missing data based on an empirical orthogonal function (EOF) decomposition, but his method could not process these images presenting extreme cloud coverage ( more than 95%), and required a long time for recon-truction..., The abnormal data in the images had a great effect on the reconstruction result. Therefore, this paper tries to improve the study result. It has reconstructed missing data sets by twice applying EOF decomposition method. Firstly, the abnormity time has been detected by an- alyzing the temporal modes of EOF decomposition, and the abnormal data have been eliminated. xcluding the abnormal data, are then analyzed by using EOF decomposition, and then the temporal modes undergo a filtering process so as to enhance the ability of reconstruct- ing the images which are no or just a little data, by using EOF. At last, this method has been applied to a large data set, ie 43 Sea Surface Temperature (SST) satellite images of the Changjiang River (Yangtze River) estuary and its adjacent areas, and the total reconstruction root mean square error (RMSE) is 0.82 ° C And it has been verified that this improved EOF reconstruction method is robust for reconstructing satellite missing data and unreliable data.