In the detection of CMM, regular surface is often a necessary feature, and the accuracy of the detection results is usually related to its sampling strategy, which includes sampling points and sampling distribution. The sampling points were studied. Because the selection of points in CMM was mostly based on experience, the detection results might not be accurate, so an algorithm of critical sampling points was proposed according to the error requirements of the features to be tested. After analyzing the disadvantages of common distribution methods, a method of low difference sequence sampling and distributing was put forward according to geometric shape of feature. Finally, the experimental results in CMM were compared. It can be proved that the proposed sampling distribution method is more reasonable.