Antimicrobial peptides have long been raised as a promising strategy to combat bacterial infection in burn wounds. Here, we attempted to rationally design small antimicrobial peptides containing unnatural amino acids by integrating in silico analysis and in vitro assay. Predictive quantitative sequence-activity models were established and validated rigorously based on a large panel of nonamer antimicrobial peptides with known antibacterial activity. The best quantitative sequence-activity model predictor was employed to guide genetic evolution of a peptide population. In the evolution procedure, a number of unnatural amino acids with desired physicochemical properties were introduced, resulting in a genetic evolution-improved population, from which seven peptide candidates with top scores, containing 1-3 unnatural amino acids, and having diverse structures were successfully identified, and their antibacterial potencies against two antibiotic-resistant bacterial strains isolated from infected burn wounds were measured using in vitro susceptibility test. Consequently, four (WL-Orn-LARKIV-NH2, ARKRWF-Dab-FL-NH2, KFI-Hag-IWR-Orn-R-NH2 and YW-Hag-R-Cit-RF-Orn-N-NH2) of the seven tested peptides were found to be more potent than reference Bac2A, the smallest naturally occurring broad spectrum antimicrobial peptide. Molecular dynamics simulations revealed that the designed peptides can fold into amphipathic helical structure that allows them to interact directly with microbial membranes.