This paper discusses predictive models for quantifying the outcome of DNA recombination employed in directed evolution experiments for the generation of novel enzymes. Specifically, predictive models are outlined for (i) tracking the DNA fragment size distribution after random fragmentation and subsequent assembly into genes of full length and (ii) estimating the fraction of the assembled full length sequences matching a given nucleotide target. Based on these quantitative models, optimization formulations are constructed which are aimed at identifying the optimal recombinatory length and parent sequences for maximizing the assembly of a sought after sequence target. Computational results show that the recombination outcome is a 'complex' function of the recombinatory length and recombined sequences and illustrate the magnitude of improvements that can be realized.