In view of the high precision requirement on the intersection point and the angle of cross laser images used in industrial applications, a sub-pixel roof-type edge detection method is proposed for cross laser images. In a serviceable area, pixel-level roof-type edge points of bright strips are firstly detected based on maximum intensity in a local region. The sub-pixel edge point is then calculated by parabola fitting of the luminance of all pixel points near a pixel-level edge point in a certain range of horizontal or vertical direction. Finally, two lines of the cross laser image are fitted by least square method, and the intersection point and the angle of horizontal line of the laser image are calculated. For 50 samples (with a 640 pixels×480 pixels camera and 0.082 7 mm/pixel as resolution of the target) at distance of 4.5 m, the standard deviation is 0.013 4 mm for x, 0.015 8 mm for y and 1.950′ for angle, much better than that of pixel-level edge points. This method is used in straightness and twist-degree inspection system for rail guides and gets satisfactory results.