With the rapid development of the optical aspheric surface industry, it has become a trend to produce aspheric lens products with a surface shape accuracy better than 0.1 μm. In the surface shape detection of the aspheric lens, due to the existence of mechanical system error, the coordinates of the detected workpiece have a deviation of 6 degrees of freedom, which will directly affect the measurement accuracy of the aspheric surface. Therefore, we need to develop an error correction algorithm with an uncertainty of only tens of nanometers for the detection system to ensure that the measurement results are closer to reality. In view of this, this paper uses data simulation to superimpose the position error and the surface shape error on the ideal aspheric surface to obtain the original three-dimensional (3D) data of the aspheric surface, and then use the revised Levenberg-Marquardt global optimization algorithm to compare the obtained original 3D data with the standard equation of aspheric surface. The principle of minimum root mean square error is used to successfully separate and correct the position error of the aspheric surface. Finally, for four glass aspheric lenses of different specifications and models, through comparing the experimental results and measurement results of the commercial aspheric profiler UA3P, a high matching result is obtained. The difference in peak-valley values is less than 5 nm, and the difference in RMS is about 0.1 nm, which verify the accuracy and robustness of the algorithm.