This paper aims to investigate facial expressions associated with visual discomfort induced by excessive screen disparities of stereoscopic three-dimensional (S3D) contents. For this purpose, we constructed a novel facial expression database regarding the visual discomfort. While viewing the realistic stereoscopic stimuli with screen disparities varying from 0 degrees to 4.66 degrees, each viewer's face was captured. The database consisted of face videos and associated comfort scores obtained by self-reporting, which might be only a publicly available database regarding the facial expressions associated with visual discomfort. Using the database, for the quantitative investigation, the facial expressions associated with visual discomfort were compared with basic emotional expressions that were well defined and universal. As a result, we observed that the emotional expression of "stressed" (i.e., anger or disgust) was highly correlated with the perceived visual discomfort (Pearson correlation coefficient: 0.91). Furthermore, the feasibility of the discomfort measurement using facial expressions obtained while viewing S3D contents was verified. Experimental results showed that the discomfort measurement using facial expression recognition could achieve a feasible performance (classification accuracy of 81.42%).