Microbes produce a variety of compounds that are used to kill or suppress other species. Traditional antibiotics have their origins in these natural products, as do many types of compounds being pursued today in the quest for new antibacterial drugs. When a potential toxin can be encoded by and exported from a species that is not harmed, the opportunity exists to use directed evolution to improve the toxin’s ability to kill other species—allelopathy. In contrast to the typical application of directed evolution, this case requires the co-culture of at least two species or strains, a host that is unharmed by the toxin plus the intended target of the toxin. We develop mathematical and computational models of this directed evolution process. Two contexts are considered, one with the toxin encoded on a plasmid and the other with the toxin encoded in a phage. The plasmid system appears to be more promising than the phage system. Crucial to both designs is the ability to co-culture two species/strains (host and target) such that the host is greatly outgrown by the target species except when the target species is killed. The results suggest that, if these initial conditions can be satisfied, directed evolution is feasible for the plasmid-based system. Screening with a plasmid-based systemmay also enable rapid improvement of a toxin.