This study focused on various approaches to rationalising the selection of test environments using on-farm trial data from 5 lentil (Lens culiniaris Medikus subsp. culinaris) genotypes. It was conducted over 3 years in 30 environments across 16 locations in Syria. There was maximum discrimination in the ratio of between-cluster to within-cluster variances, based on genotype yield responses to the environments. Four clusters represented the test locations, reflecting a gradient in the levels of yield and seasonal rainfall. We observed significant genotypic differences and genotype × environment interactions. Genotype × cluster interaction accounted for a substantial portion of the genotype × environment interaction. This supported a reduction in the number of test locations to evaluate genotype and environment interaction. Temporal interactions were either low or insignificant. The improved lines produced stable and significantly higher yields than the local cultivar. The structure of the clusters formed indicated the presence of research stations in each cluster. We recommend that locations for future on-farm testing should include one research location and a farmer field in each cluster (or the mega-zone) so formed. Climatic variables or geographical nearness cannot replace the role of genotype response when rationalising test locations.