Little is known about the effect of farmer selection on abiotic stress tolerance of maize. A set of maize varieties was evaluated in a field trial during the 2003/04 season at Harare to determine the pattern of phenotypic diversity and classify the varieties in groups, to further evaluate representative sets of landraces that originate from different environments. The set comprised original open-pollinated varieties (OPVs) introduced into Zimbabwe, Zambia and Malawi from the USA, local landraces collected from smallholder farmers in the three southern African countries, historically important OPVs, and improved varieties developed in the region. Adjusted means for anthesis date, number of kernel rows per cob, 100-kernel weight, kernel length, kernel width, plant height, ear height, ear length, ear diameter, rachis diameter, cob diameter and stalk circumference were used in classifying the varieties. Three distinct, non-overlapping groups were obtained using cluster analysis for the 294 varieties. Group 1 was composed entirely of local landraces characterized by the lowest grain yields among the three groups at 3.395 Mg ha–1. Number of days from planting to 50% anthesis was 75, number of ears per plant 0.76, 100-kernel weight 41.14 g and number of kernel rows per cob 11. Group 2 was composed mostly of local landraces that have phenotypic characteristics similar to the OPV Hickory King, historically important OPVs from Zimbabwe (Salisbury White, Southern Cross), and Hickory King from the USA. The fact that the historically important OPVs from Zimbabwe clustered with Hickory King from the USA may indicate a close relationship or common parentage among them. This group is characterized by intermediate grain yields at 4.388 Mg ha–1, 75 days from planting to 50% anthesis, 0.77 ears per plant, 50.0 g per 100 kernels, and 10 kernel rows per cob. Group 3 was composed of improved varieties from the seed companies in the three countries, most of the introduced OPVs from the USA, OPVs from the International Maize and Wheat Improvement Center (CIMMYT) and the Zambia National Program, and a few local landraces. This group had the highest average grain yield at 4.932 Mg ha–1, number of ears per plant at 0.88 and number of kernel rows per cob at 13. Number of days to 50% anthesis at 72 and 100-kernel weight at 38.21 g were the lowest among the three groups. The fact that most of the improved varieties clustered separately from the landraces indicates that breeders but not smallholder farmers have been selecting for different morpho-agronomic traits. The effect of breeders’ selection is evident in the high grain yields, early maturity, high number of kernel rows on ears and high number of ears per plant. From this study, it can also be hypothesized that the landraces that grouped with improved and introduced varieties (group 3) may actually be ‘creolized’ varieties, that is, hybridizations between traditional landraces from groups 1 and 2 with improved varieties available in the three countries. From these groupings, the upper 25% of the varieties, representing those with the highest average diversity of the clusters, were used to form a subset for further evaluation under different stress levels.