Three different sources of dwarfism Tordo, Vican-71 and Olesen were used in a dialell cross. Parents, F1's, F2's and reciprocal backcrosses were tested in an experiment using pots, carried out at the Experimental Center of Campinas, State of São Paulo, Brazil. The experimental data for plant height, head length, number of spikelets per spike, number of grains per spike, 100-grain-weight, number of grains per spikelet, number of spikes per plant and grains yield were determined on a individual plant basis. Partial dominance was observed for short plants in F1 generation when the cultivar Tordo was crossed with Vican-71 and a recessive behaviour of the genes of Olesen when this cultivar was crossed with Vican-71. Broad sense heritability estimates were high for: number of spikelets per spike, number of grains per spike and number of spikes per plant considering the cross Tordo x Vican-71; for head length, number of spikelets per spike, number of grains per spike and per spikelet, number of spikes per plant and grains yield considering the cross Tordo x Olesen; and for head length, number of spikelets per spike, number of grains per spike and per spikelet, 100-grain-weight and number of spikes per plant considering the cross Vican-71 x Olesen. Considering the cross Vican-71 x Olesen the narrow sense heritability estimates for head length, number of spikelets per spike, number of grains per spike, 100-grain-weight, number of spikes per plant and plant height presented high values, suggesting that selections for these characters would be effective in the F2 or F3 populations. For the other characters in the cross Vican-71 x Olesen, and for all characters in the crosses Tordo x Vican-71 and Tordo x Olesen, the narrow sense heritability values were medium or low, suggesting that selection would be effective in the later generations where genetic value of the progenies would be more precisely determined. The phenotypic correlations between plant height and all the studied agronomic characteristics (considering all crosses) were positive and significant except for the correlation between plant height and 100-grain-weight of the cross Tordo x Vican-71. The results indicated that the studied sources of dwarfism could be used alone or combined in a cross without causing any undesirable association; as long as a large F2 population be available in order to provide adequate frequency of desirable recombinants.