Plant breeders select at specific sites to achieve a response at other locations within the target population of environments (TPE). The effects of this selection can be modeled using the theory of correlated response. The value of a test location is determined both by the genetic correlation of genotypic effects at that location with those in the TPE as a whole (rG) and by the precision, or broad-sense heritability (H) with which effects are measured at that location. Together, these parameters can be used to predict gains in the TPE from selection at specific test sites, using multi-environment trial (MET) data sets. This method of determining the predictive value of trial locations is superior to widely-used methods such as GGE biplot analysis in that it integrates both the discriminatory power of a location and its correlation with the TPE into a prediction of selection response, facilitating quantitative comparisons among the value of test locations. The method was used with historical data sets from the CIMMYT tropical maize breeding program to compare test locations with respect to the indirect selection response they are predicted to generate in Asian, Central, and South American TPE. Takfa, Thailand, generated a high predicted response to selection for grain yield at Indian and Southeast Asian locations, and a moderate response in Central and South American locations. CIMMYT's main tropical lowland breeding site in Agua Fria, Mexico, generated a high predicted response in the Central American TPE, but has very limited predictive value for the Asian TPE