THE text-book treatment of linear curves of best fit purports to give the best estimate of the relation w1 = pv1 between variables of true value w1, v1, from a set of measured values, w, v. It is applicable when only one parameter is subject to errors of measurement. Thus, when only the values of w are subject to error, the best estimate is given by the curve which minimizes the sum of squares of deviation of w, that is, the regression line of w on v ; and for errors in v only, the regression line of v on w should be taken.