The usual way of judging the quality of data in a least squares calculation is wrong; a better way is suggested. In the procedure of averaging results of experiments on the same difference of atomic masses, the errors may nor be corrected by a consistency factor, as is commonly done.