Regressions of yields of cultivars upon means of sets of cultivars over diverse environments are often used as measures of stability/adaptability. Prolonged selection for performance in environments of high yield potential has generally led to unconscious selection for high regressions. If adaptation to poor environments is required (as it often is in Third World agriculture), common sense suggests that low regressions could be exploited for the purpose. Simulations show that systematic selection in the poor environment is required, not merely trials of potential cultivars after selection in a good environment. In effect, systematic exploitation of a GE interactions effect is proposed. The effects are large enough to reduce correlated responses in different environments to zero. Orderly experimental studies are needed but not available. What information there is does not disagree with the theory developed here.