Efficiency of hybrid breeding could be greatly enhanced by reliable methods of prediction of F1 hybrids values or heterosis. Use of parents per se values through different approaches (including distance computation) should be of poor efficiency for species that exhibit high heterotic effects, such as Maize. In such a situation, testcross designs lead to efficient predictions of Fl hybrids values, as was illustrated by two experimental studies on Maize. When using a classical approach based on general combining ability (GCA) estimation, the most important discrepancies between predicted and actual values of the hybrids were obtained for overestimated crosses (actual value poorer than predicted). It was demonstrated, both theoretically and experimentally, that the use of parents * testers specific effects (SCA) could enhance the prediction efficiency without additional measurements. However this method still leads to misleading conclusions for some hybrids between ”original” lines (poorly related to any tester). RFLP assays should provide a powerful way to detect such situations through the estimation of parents similarity and a better knowledge of germplasm structuration.