The mean yield in grams of dry matter per plant was measured under two cutting regimes, frequent (F) and infrequent (R). The results could not be fitted adequately by linear regression of the mean yield (y) on the number of plants omitted or substituted (x). Curvilinear regression lines of the form y = a + bx2 or y = a + b1x + b2x2 were successful in fitting the data, but they led to unacceptable predictions when they were extended from substitution experiments, where x is negative, to addition experiments in which x is positive. When a square root transformation was used, linear regressions showed an adequate fit with the observations. Furthermore the heterogeneity of the duplicate error variance, found when y itself was employed, was greatly reduced, and the mean square deviations of the observations from the regression lines were homogeneous within the two cutting regimes. The square root transformation also avoids the grossly unacceptable features of polynomial regressions when they are extended to predict the outcome of addition experiments.The regression coefficients were used to derive estimates of the competitive abilities of the three genotypes in both monocultures (intra-genotypic competition) and duocultures (inter-genotype competition). These competitive values differ between the two cutting regimes, being lower under the F regime than under the R.