This paper investigates 3 statistical techniques often employed in comparing 2 growth curves. It shows: (1) that a polynomial regression model, when justified, is superior to a comparison based on the means only; (2) that a polynomial model, when justified, is superior to a nonparametric method introduced by Rao for comparing 2 or more growth curves; (3) that the type I errors arising in using the classical F distribution in testing a set of regression coef-ficents when correlation exists between observations are very large even for a modest amount of correlation.