In a recent article appearing in this journal Turner and Stevens [1959] have presented an excellent discussion of regression analysis of causal paths based on the early work of Sewall Wright and others. authors note that considerable work on the treatment of multiple equations has also been done by econometricians. (See the papers by Haavelmo [1943, 1944], Koopmans and Reiers0l 1950].) purpose of the present note is to make known to biologists and biometricians several methods of genieralized linear estimation of structural parameters not yet widely known outside of econometrics, in particular the leastvariance difference method. latter method, in two different forms, has been introduced into the econometric literature by Theil [1953] and the present author [1957]; but the principle on which it is based and, indeed, the mathematical technique itself have been known for some considerable time in the Theory of Errors, the principle having been laid down and the technique fully adumbrated by Gauss [182123j. This principle, in Gauss's words, asserts that The most plausible value of a quantity that is a given function of the unknown quantities (i.e. the parameters) of the problem is found by one's substituting for the latter their most plausible values as determined by the Method of Least Squares. (Gauss [1823], pp. 34ff., pp. 102-3.) This principle foreshadows the well-known modern theorem about the invariance of maximum-likelihood estimators. technique of least-variance difference estimation can be formulated exactly as Aitken's [1935] generalization of Gauss's method of treating independent observations of unequal precision (Gauss [1809], p. 214) to the case of interdependent observations. plan of this note is as follows: Section 2 contains a discussion of some logical features of causal