The potential of bambara groundnut as a crop-based approach for the sustainable reduction of protein–energy malnutrition prevalent in Nigeria is still being explored. However, one of the limitations to sustainable production of this adapted species is low and unstable yield. Here, we employed multiple statistical analyses to determine traits that could be useful in predicting grain yield. Furthermore, additive main effect and multiplicative interaction and genotype × genotype × environment models were used to identify high yielding and stable landraces. Twenty-four bambara landraces were evaluated in two contrasting dry land growing seasons at four environments. Based on the results of correlation, path coefficient values and stepwise multiple regression analysis, it is reasonable to assume that seed growth rate would be effective as indirect selection criteria for grain yield improvement in this set of breeding materials under early and late planting environment. However, optimizing biomass growth rate, pod fill period, 100-seed weight, number of pods per plant and time to flowering may have important role in the improvement of grain yield in bambara groundnut. Due consideration should be given to 100-seed weight and time to flowering in the early dry and late planting environment. The possibility of successful cultivation of bambara groundnut in forest–savannah transition agricultural zone of Nigeria, particularly when planting is done around early August has been established in this study. However, based on the strong association between seed yield and its related characters in early and late planting environments, selection for higher seed growth rate combined with large number of pods appears to be the best indirect traits when selecting bambara groundnut plants to enhance grain yield under early and late planting environments. While landraces TVSu 1520 and 1578 were considered the best adapted landraces, TVSu 1670 and 1518 are promising landraces that could contribute to increasing grain yield stability.