For clear binocular vision in the brain, the neural interaction for accommodation and vergence is performed via inter-operation of two cross-links: accommodative-vergence (AV) and vergence-accommodation (VA). However, when people watch stereo images on stereoscopic display, the neural operation is interrupted due to alternations of the cross-link gains, which are attributed as the main reason of visual discomfort on stereo images in terms of Accommodation-Vergence Mismatch (AVM). In this paper, we present a novel visual discomfort prediction algorithm dubbed 3D-AVM Predictor to quantify the visual discomfort by including neural responses to sensory stimuli of both accommodation and vergence. In particular, we define the 3D local bandwidth (BW) based on physiological optics of binocular vision and foveation for optical activity of accommodation from the perspective of the VA cross-link. Since the 3D-AVM Predictor encompasses the anomalous motor responses of both accommodation and vergence, it is demonstrated that the performance is statistically superior to those of conventional works which rely on the factor of disparity distribution only.