Seed yield is of little agronomic importance in alfalfa (lucerne) but is critical in the marketing of varieties. In order to develop breeding criteria, the quantitative genetics of seed yield components must be studied. The objective of this study was (1) to evaluate combining ability for seed yield and its components and (2) to estimate the genetic control of these same characters. A 7 x 7 diallel mating design, including reciprocals, among parents from different varieties, and a 7 x 7 factorial design within the 'Flamande' population were evaluated. In the diallel design, where the progenies were evaluated over two growing seasons, a high genotype effect was found for seed yield and its components while genotype x year interaction was only significant for seed yield per plant. General combining ability (GCA) effects explained most of the variation due to genotype effect. Specific combining ability (SCA) and reciprocal effects were only significant for seed yield per plant. 'Europe-1' 'Rival-5' and 'Medalfa-7' were the most promising parents, conferring the highest GCA for most of the characters evaluated, especially for seed weight per inflorescence. In the factorial design, seed weight per pod was the only character for which the year effect was not significant. Male and female effects were significant for all characters and these effects were larger than the male x female interaction variance for all the characters. These results were stable over the 2 years for most characters. Additive variance was larger than dominance variance, for all traits. This resulted in high narrow-sense heritabilities, especially for seed yield per plant, seed weight per inflorescence and number of seeds per pod. Seed weight per inflorescence and number of seeds per inflorescence were highly correlated with seed yield per plant among the full-sib (FS) families, in both experiments. Increase in seed yield potential in alfalfa could be achieved by the use of seed weight per inflorescence as a selection criterion.