This paper presents a rigorous mathematical formulation for evaluating error behaviors and reliabilities in the computation of vanishing points for sensing 3-D from images. First, an optimal vanishing point estimation scheme is constructed by adopting a simple statistical model which assumes that image errors result from edge fitting in digital images. Then, reliability of thus computed optimal, estimate is theoretically evaluated, and the confidence level of the estimate is statistically deduced in quantitative terms. Finally, a simple scheme of calibrating the camera focal length is constructed in such a way that the reliability of the computation is maximized.