A three-dimensional visual sensor is disclosed. A two-dimensional image of a two-dimensional feature portion (15) including a point (11) determined on a work (13) is acquired, and N (N≧2) points are determined. A slit of light is projected by a projector (1), an image of a projected portion (19) is obtained, and M (M≧2) points are determined. The three-dimensional position of the intersection point between each straight line connecting the N points and a point in a camera (2) and the slit of light plane is determined on the sensor coordinate system, and transformed to the data on the reference coordinate system (on or parallel to the slit plane). The three-dimensional positions of M points are similarly subjected to coordinate transform to the data on the reference coordinate system. Straight lines (or curves) defined by the M and N points are determined, respectively, by the least squares method. The intersection point between the two straight lines is determined and returned to the data on the sensor coordinate system by inverse transform. The template matching method may alternatively be used for detecting the feature portion. As a result, the measurement accuracy and the reliability are improved for the three-dimensional visual sensor for measuring the three-dimensional position of a specific point by radiating the a slit (or a pseudo-slit) of light on the surface of the object.