The model of Gardner and Eberhart (Biometrics 22: 439-452, 1966) was u
sed for the analysis of variety diallel crosses, in which the F-2 gene
rations, instead of the F-1 hybrids, are used. The least square proced
ure was used to derive the formulas for estimation of effects (compone
nts of means) and for the analysis of variance. The estimators of the
mean (mu) and variety effects (v(i)) did not differ from the original
model; but the estimators of the heterosis components for F-2 data wer
e twice those obtained from the original model, In the same way, the s
tandard errors of the estimates were different only for the heterosis
components, when compared with formulas from the original model, while
the formulas for the analysis of variance were the same.