The ordinary major axis of a bivariate normal distribution is unsuitable for describing the functional relationship (in Lindley’s general sense) between two naturally variable quantities, because it is not invariant with scale and because qualitatively unlike quantities are subtracted in computing it. The bivariate structural relationship is unsuitable because one of the parameters required, λ, cannot be estimated from naturally variable data; in fact λ has no objective meaning for such data. The standard major axis (or GM regression line) has neither of these defects, and it is the line that minimizes the sum of products of the absolute vertical and horizontal deviations of data points from itself.