This paper proposes object-dependent disparity features to predict the visual discomfort in stereoscopic 3-D images. The proposed object-dependent disparity features quantify the level of visual comfort influenced by disparity gradient of nearby objects and object width, respectively. They consist of relative disparity (mean of disparity difference between nearby objects) and object thickness (ratio of mean width to mean absolute disparity of an object). The prediction performance of the proposed disparity features is evaluated using various types of stereoscopic images. Experimental results demonstrate that the combined use of the proposed object-dependent disparity features substantially improve the prediction performance of the conventional disparity magnitude-and spatial complexity-related features. The performance gain ranges from 0.045 to 0.135 of correlation coefficient, compared with the feature combinations used in the conventional visual comfort metrics.Index Terms-Depth object, disparity feature, stereoscopic image, visual discomfort prediction.