Policies aimed at alleviating the growing problem of drug–resistant pathogens by restricting antimicrobial usage implicitly assume that resistance reduces the Darwinian fitness of pathogens in the absence of drugs. While fitness costs have been demonstrated for bacteria and viruses resistant to some chemotherapeutic agents, these costs are anticipated to decline during subsequent evolution. This has recently been observed in pathogens as diverse as HIV and Escherichia coli. Here we present evidence that these genetic adaptations to the costs of resistance can virtually preclude resistant lineages from reverting to sensitivity. We show that second site mutations which compensate for the substantial (14 and 18 % per generation) fitness costs of streptomycin resistant (rpsL) mutations in E. coli create a genetic background in which streptomycin–sensitive, rpsL+alleles have a 4–30 % per generation selective disadvantage relative to adapted, resistant strains. We also present evidence that similar compensatory mutations have been fixed in long–term streptomycin–resistant laboratory strains of E. coli and may account for the persistence of rpsL streptomycin resistance in populations maintained for more than 10 000 generations in the absence of the antibiotic. We discuss the public health implications of these and other experimental results that question whether the more prudent use of antimicrobial chemotherapy will lead to declines in the incidence of drug‐resistant pathogenic microbes.