Homologous recombination is a technique used for performing precise genomic manipulations, and this makes it potentially ideal for gene therapy. The rate of spontaneous homologous recombination in human cells has been too low to be used experimentally or therapeutically but, by inducing a DNA double-strand break (DSB) in the target gene this rate can be stimulated. Zinc finger nucleases (ZFNs) are synthetic fusion proteins that can induce DSBs at specific sequences of DNA and stimulate gene targeting. Although the success of ZFNs in this application has been demonstrated, several issues remain. First, an optimal, generalized method of making effective and safe ZFNs needs to be determined. Second, a systematic method of evaluating the efficiency and safety of ZFNs is needed. We compared the gene-targeting efficiencies and cytotoxicity of ZFNs made using modular-assembly and ZFNs made using a bacterial 2-hybrid (B2H) selection-based method, in each case targeting the same single site. We found that a ZFN pair made using the B2H strategy is more efficient at stimulating gene targeting and less toxic than a pair made using modular-assembly. We demonstrate that a pair of three-finger B2H ZFNs is as efficient at stimulating gene targeting as ZFNs with more fingers, and induce similar or lower rates of toxicity.