One promising way to breach the skin's natural barrier to drugs is by the application of chemicals called penetration enhancers. However, identifying potential enhancers is difficult and time consuming. We have developed a virtual screening algorithm for generating potential chemical penetration enhancers (CPEs) by integrating nonlinear, theory-based quantitative structure-property relationship models, genetic algorithms, and neural networks. Our newly developed algorithm was used to identify seven potential CPE molecular structures. These chemical enhancers were tested for their toxicity on (a) mouse embryonic fibroblasts (MEFs) with MTT assay, and (b) porcine abdominal skin by histology using H/E staining at the end of a 48-h exposure period to the chemicals. Further, melatonin permeability in the presence of the enhancers was tested using porcine skin and Franz diffusion cells. Careful toxicity tests showed that four of the seven "general" CPEs were nontoxic candidate enhancers (menthone, 1-(1-adamantyl)-2-pyrrolidinone, R(+)-3-amino-1-hydroxy-2-pyrrolidinone, and 1-(4-nitro-phenyl)-pyrrolidine-2,5-dione). Further testing of these four molecules as potential melatonin-specific CPEs revealed that only menthone and 1-dodecyl-2-pyrrolidinone provided sufficient enhancement of the melatonin permeation. The results from our permeability and toxicity measurements provide validation of the efficacy and ability of our virtual screening algorithm for generating potential chemical enhancer structures by virtual screening algorithms, in addition to providing additional experimental data to the body of knowledge.