To solve the three-dimensional(3-D) position calibration problem of spatial feature points, a vision calibration method of spatial point position under multi-view scene determination is proposed. Based on the computer vision principle, the scene pose of image is determined through placing a chessboard pattern in the scene, and then the 3-D position of spatial point is solved by using the image coordinates of spatial point in multi-view images and the scene poses of multi-view images. Results show that the 3-D position of spatial point calibrated by the method has a distance error less than 1 mm, which means the method is suitable for high-accuracy point position calibration.