The generation of a good library is critical for the success of directed evolution. In this presentation, we will suggest improved directed evolution methodologies using computational and experimental strategies developed from sequencing and phenotypic results obtained from the recombination of fluorescent proteins. We will demonstrate that DNA-shuffling and recombination PCR have intrinsic advantages, disadvantages, and biases depending on the number and type of genes recombined. When recombining two genes, DNA-shuffling can have even and odd number of crossovers but it is a skill-intensive protocol. Recombination PCR is easy to apply and it can recombine genes with lower identity level without parental background but it is template sensitive. The recombination of genes can also be improved by the use of computational tools such as support vector machines to identify interacting positions and to reduce the number of inactive variants resulting from recombination of sensitive regions.