In this paper,we propose a new m ethod for stereoscopic image quality assessment based on deep feature learning.Different from traditional stereoscop ic image quality evaluation methods,we use convolutional neural network (CNN) to extract features,and the quality prediction process can be divided into two steps:training and testing.I n the training phase,the images are divided into blocks to train the CNN network and then the block features can be extracted by the CNN.Through differe nt integration ways,global features of images are calculated from block features.We establish the regres sion model between subjective scoring values and global features by support vector regression (SVR).In the tes t phase,quality scores of left,right and cyclopean images can be predicted according to the trained CNN network and regression model.Finally, the qualities of left,right and cyclopean images are combined together to get the stereoscopic image quality.Ex perimental results on the publicly available LIVE-I and LIVE-II data sets demonstrate the superiority of the prop osed method.The spearman rankorder correlation coefficient (SROCC) scores of our method in the LIVE-I and LIVE-II reach 0.94and 0.89respectively.The eva luation results are accurate and consistent with the subjective perception of the human eyes.