Aiming at the evaluation problem of symmetric distortion and asymmetric distortion image, a non-reference stereo image quality evaluation method based on binocular fusion is proposed. Firstly, the left and right viewpoint images of the stereo image are decomposed into Laplacian pyramid sequences respectively, and the fusion coefficients of each layer are determined by using the image ave- rage gradient and the region energy. On the basis of the binocular weighted model, the two sequences are merged layer by layer and the cyclopean image is reconstructed. Then, the multi-scale, multi-directional frequency domain transform features and the contrast, entropy, energy and inverse difference moment features of the left and right viewpoint images and the cyclopean images are extracted. Finally, feature parameters are trained as input to the support vector regression model and the image quality is predicted. The correlation analysis is performed under LIVE 3D phase I and LIVE 3D phase II image databases. The Pearson linear correlation coefficient and Spearman rank correlation coefficient reach 0.96 and 0.95 respectively. The results show that the prediction results of stereo image quality have higher consistency with subjective evaluation values.