Least　squares　association　of　geometrical　features　plays　an　important　role　in　geometrical　product　specification　and　verification.　Most　existing　algorithms　for　the　least　squares　association　today　usually　do　not　give　the　covariance　matrix　associated　with　the　parameters　of　the　respective　geometrical　feature.　The　reason　is　that　the　complexity　of　these　algorithms　can　be　very　high,　because　partial　differential　quotients　are　needed.　If　the　necessary　partial　difference　quotients　are　calculated　by　hand　and　subsequently　coded　into　an　algorithm,　there　is　a　high　risk　to　introduce　unwillingly　errors.　This　paper　shows　how　the　least　squares　algorithm　can　automatically　be　generated　solely　from　the　equation　specifying　the　distance　function　of　the　measured　points　from　the　geometrical　feature.