Image edge detection process is problematic in that noise suppression and detail retention can not be taken into account, this prompted us to propose an edge detection algorithm based on Bertrand surface model. On the basis of determined pixel edge,selecting strip domain along edge direction as fitting area,with the Bertrand surface characteristic of normal lines at various points along generatrix line are in one plane, transforming the pixel information in fitting surface area into active coordinate of edge curve,and fitting coordinate and normalized gray value to obtain normal distance between sub-pixel edge and pixel edge for sub-pixel edge detection.Adopt the vision measurement system to experiment with the gauge block line edge,andcompared to the improved algorithm of sub-pixel edge detection based on Facet surface fitting, the results show that algorithm of edge detection based on Bertrand surface model has high location accuracy, the first grade gauge blocklinear error is within 1 μm, and the multiple measurement error is －0.811 μm,it also has high reliability. The oil pump body measurement demonstratesthat this algorithm can be applied to precision measurement of mechanical parts, especially suitable for measurement of center distance, bore diameter, and so on.