Maintaining quinolone concentrations outside the mutant selection window (MSW) between the MIC and mutant prevention concentration (MPC) was suggested by in vitro and in vivo studies to prevent the selection of resistant mutants. However, selection also may depend on the presence of resistant bacterial mutants at the start of treatment, which is highly dependent on the initial inoculum size. In this study, a mouse thigh bacterial infection model was used to test the influence of different exposures to marbofloxacin on the selection of resistant bacteria after infection with a low (10 5 CFU) or high (10 8 CFU) initial inoculum of Escherichia coli. The inoculum size was shown to influence the exposure to marbofloxacin and the values of pharmacokinetic/ pharmacodynamic indices. When the abilities of the indices time within the MSW (T MSW ), area under the concentration-time curve of 0 to 24 h divided by the MIC, and the maximum concentration of drug in plasma divided by the MIC to predict the selection of resistant bacteria were compared, only T MSW appeared to be a good predictor of the prevention of resistance for values less than 30%. When the T MSW was higher than 34%, the selection of resistant bacteria occurred less often in thighs initially infected with the low inoculum (11/24; 46%) than in those infected with the high inoculum (30/36; 80%), suggesting that the selection of resistant mutants depends on both the T MSW and inoculum size. The relevance of these results merits further investigation to test different strategies of antibiotic therapy depending on the expected bacterial burden at the infectious site.Resistances to fluoroquinolones can occur spontaneously in bacterial populations at a frequency of about 10 Ϫ6 to 10 Ϫ8 (5) by following a stepwise process that involves mutations in genes coding for the targets DNA gyrase and topoisomerase IV (23, 29). Consequently, if the bacterial load at the infectious site exceeds the inverse of the mutation frequency, it can be presumed that a small resistant subpopulation already coexists with a larger susceptible population before any antimicrobial treatment is administered. Traditionally, in vitro antimicrobial studies and animal infection models have been used to assess the reduction in the total bacterial population at an infectious site while often ignoring the impact of drug pressure on the amplification of the drug-resistant subpopulation (2,8). Thus, the values of pharmacokinetic/pharmacodynamic (PK/PD) indices determined from these experiments were selected previously to predict the bacteria killing and not the selection of resistant bacteria.