A breeding and selection technique for improvement of commercial hybrids in diploid organisms has been outlined, and designated as recurrent reciprocal selection. Theoretical comparisons have been made of the limits of improvement and improvement rates to be expected of this method, selection for general combining ability, and the method proposed by Hull (4). They indicate that under no circumstances would reciprocal selection be more than slightly inferior to the better of the other two. However, it would be definitely superior selection for general combining ability for loci at which there is over-dominance or if a situation analogous to that with over-dominance exists due to linkage; and it would be definitely superior to the method proposed by Hull for loci at which there is partial dominance. In view of the reality of linkage and the improbability of partially dominant genes being of minor importance in real genetic situations, there is good reason to believe reciprocal selection will in practice be more effective than the other methods discussed in this paper.