Analysis of a conventional diallel cross offers only perfect-fit estimates of the genetic components of variation, but no test for the goodness of fit of the model based on these estimates. When F2 progenies are available, however, combining F1 and F2 diallels in a single experiment overcomes these problems. Least-squares estimates of these components can be calculated, errors attached to them and the goodness of fit of the resultant model tested. This analysis was applied to data on the severity of yellow rust infection in an F1/F2 half-diallel cross among eight bread wheat lines adapted to the East African highlands. After removing two interacting arrays, genetic analysis indicated that an additive/dominance model of gene action satisfactorily explained the variation observed among the remaining six parents and their progenies, in both the individual F1 and F2 diallels and the combined F1/F2 diallel. Resistance to yellow rust was dominant to susceptibility and genes for increased resistance were more frequent.