With the increasing necessities for reliable printed circuit board (PCB) product, there has been a considerable demand for high-speed and high-precision vision positioning systems. To locate a PCB board with high accuracy and reliability with the positioning mark images, a new visual positioning method is introduced. Considering the limitations of Lyvers subpixel edge detection algorithm, such as the lower positioning accuracy, the larger errors at the neighborhood of the edge intersections, and the more computing time and the more cumbersome process of threshold selection, we propose an improved algorithm, in which coarse and accurate edge detection methods are adopted, and a new criterion is put forward to detect the edge points with Otsu method. Furthermore, a formula is developed to determine the edge intersections, whose subpixel coordinates are calculated with bilinear interpolation and conjugate gradient method. Additionally, the causes of principle errors are explored, and the error compensation formulas are derived at any angle edges of the three-level edge model. The last experimental results show that the improved algorithm has good versatility, and compared with Lyvers and Ghosal algorithm, the process of threshold selection is easier, the detected edges are thinner, and the positioning accuracy for all edge points is much higher.