Selection for specific adaptation in a breeding programme may be required when significant genotype×environment (G×E) interaction occurs within a target production region. Knowledge of the key stress factors underlying the G×E interaction permits delineating a production region into homogeneous zones for the purpose of recommending specific varieties. Twenty-eight cowpea genotypes were evaluated in 2002 and 2003 at four locations in the Guinea and Sudan savannah zones of Ghana, to assess grain yield stability and its relationship with intensity of stress in each production environment. Data were analysed in a year-location cross-classification, producing eight environments. Significant variations were observed between the genotypes for grain yield and grain yield components, biomass production, days to maturity, and resistance to Striga, pod-sucking bugs, and bacterial blight severity. Across environments, stress intensity (in terms of grain yield) was negatively correlated with the genotypic and error variances but positively correlated with the phenotypic, genotypic and experimental coefficients of variation. Stress intensity was also negatively correlated with heritability for grain yield. Genotypic correlations for grain yield between locations were moderate to high. The b-value as a measure of yield stability showed positive correlations with mean yield over all environments, and also with average yield in each of the three most favourable locations. In the least favourable location, the b-value showed a weak inverse relationship (P=0·10) with grain yield. Although direct selection for grain yield was considered superior in the target environments, simultaneous selection for grain yield and tolerance to stress in the less favourable environments will identify genotypes broadly adapted to the major cowpea growing regions.