In this paper, we propose a new visual discomfort prediction method for stereoscopic 3D contents. Our features are computed from stereoscopic 3D video such as disparity, motion, contrast, spatial complexity of salient objects and brightness and binocular asymmetries degree between left and right image in a 3D scene. The salient object is detected by region based multimodal information such as color, disparity and location of regions. The visual discomfort is estimated by prediction function trained by Support Vector Regression (SVR) method with temporal pooling strategy. The experimental results showed that the proposed method shows good prediction results correlated with subjective assessment results. In addition, we compare the error in visual discomfort measurements between the subjective test and our method.