A localization method of robot local polishing was proposed.Points of seed were first selected from measured data points obtained collectively,and the eigenvalue coordinates of these points,which were invariant under the body transformations,were constructed using the covariance matrix in statistical analysis.Then,a list of corresponding points on the model surface was generated by directly searching the same eigenvalue coordinate on the model surface.Subsequently,the principal axes corresponding to eigenvalue were normalized and a method of calculating the transformation matrix was proposed.On the basis of these,the proper localization transformation was achieved by minimizing the distance target function with solution to the matrix.Finally,an experimental result gives that this method is simple,reliable and easy to implement,which is especially suitable in localization of local polishing of parts with complex surfaces.