Thirty three families obtained in biparental and polycrosses were evaluated at two sites in the State of São Paulo - Brazil, Piracicaba and Jaú, in order to quantify the Family x Environment (FE) interaction and its effect on the expected response to selection. Stalk height, stalk diameter, number of stalks, sugar content (Brix), cane yield per hectare (TCH) and sugar yield per hectare (TB) were evaluated. Results indicate a significant FE interaction (p<0.01) for all traits measured. Although always less than the family variance (<img src="../../../../img/revistas/brag/v61n1/o_circ2f.gif" align=absmiddle>), the FE interaction variance (<img src="../../../../img/revistas/brag/v61n1/o_circ2fe.gif" align=absmiddle>) was from 12,3% to 75,7% of the family variance (<img src="../../../../img/revistas/brag/v61n1/o_circ2f.gif" align=absmiddle>) for stalk diameter and TBH, respectively. The decomposition of the FE interaction variance, in genotypic and phenotypic components, made evident its complex nature, making it difficult to select families adapted to both sites. Simulated responses to selection (SRf) showed that indirect selection (selection in one environment and expected response in another) has very low efficiency and was in some cases, such as TBH, negative. Selection based on the mean of both sites presented gains, however these gains would have been 62% and 43% greater (for Piracicaba and Jaú, respectively) had the selection and response been made on the same site.
sugarcane; family x environment interaction; response to selection
