Utilization of heterosis effects in breeding soybean (Glycine max (L.) Merr) cultivars for commercial production is still a possibility. One objective of this research was to determine heterosis expression in soybean germplasms obtained from the crossing of parents selected on the basis of restriction fragment length polymorphism (RFLP) diversity and isozyme loci (ISO) constitution compared to parents selected for high-yielding (HY) ability and differences in geographic origin (GO). The second objective was to compare mean heterotic performance of F[subscript]1 hybrids and F[subscript]2 bulks evaluated in hill- and row-plots;Twenty-one genotypes of each of Maturity Groups (MG) II and III, selected according to four criteria: HY, and for differences in RFLP, ISO, and GO were used in crossing to obtain 24 F[subscript]1 hybrids and their F[subscript]2 bulks. Parents, F[subscript]1 hybrids, and F[subscript]2 bulks grown in hill plots and parents and F[subscript]2 bulks grown in row plots were evaluated for seed yield (YLD), days to physiological maturity (MAT), lodging (LDG), height (HT), days to flowering (FLW), seed weight (SDWT), and protein (PROT) and oil (OIL) content in 1990 and 1991. Midparent (MPH) and high-parent (HPH) heterosis for the F[subscript]1 and F[subscript]2 generations were calculated for each hybrid combination. RFLP-based genetic distances among the parents of each hybrid were obtained and correlated with heterotic effects for YLD. Phenotypic and rank correlation coefficients for YLD were calculated over all criteria between all possible combinations of generations and plot types;The results indicated that of the two criteria at the molecular level, selecting parents for different isozyme constitution was more advantageous for the expression of heterosis in seed yield than the use of RFLP differences. The results observed in the study suggest that in soybean, the measure of genetic distances calculated from RFLP data may not be effective in establishing the genetic information required for the prediction of hybrid performance;The evaluations conducted in the F[subscript]1 and F[subscript]2 generations in hill and row plots of both MGs, suggested that midparent heterotic effects determined at the F[subscript]2 generation in hills could be of value to predict the F[subscript]1 hybrid heterotic performance if they also were planted in hill plots. There was no association between the heterotic expression evaluated in the F[subscript]2 generation tested in rows and in the F[subscript]1 generation planted in hills. (Abstract shortened by UMI.)