Device-related infections are difficult to treat with antibiotics alone. Standard susceptibility tests do not correlate with treatment success. Therefore, the utility of a pharmacokinetic in vitro model has been evaluated in comparison with the tissue-cage infection model in guinea pigs. The bactericidal activity of 28 treatment regimens has been studied by using three different test strains. In vitro efficacy was defined as reduction in the number of suspended or adherent bacteria, and in vivo efficacy was defined as reduction in the number of bacteria in tissue-cage fluid. Test results between the two models (in vivo and in vitro) correlated well, with correlation coefficients of 0.85 for in vivo efficacy versus in vitro efficacy against suspended bacteria and 0.72 for in vivo efficacy versus in vitro efficacy against adherent bacteria (P < 0.05) for Staphylococcus aureus, 0.96 and 0.82 (P < 0.05) for Staphylococcus epidermidis, and 0.89 and 0.97 for Escherichia coli, respectively. In contrast, standard susceptibility tests, ratios of MICs to trough or peak levels, ratios of the area under the curve to the MIC, or time above the MIC were not predictive for therapeutic outcome in either the in vitro or in vivo model. In both models, the bactericidal activity levels with combination regimens were significantly higher than those with single-drug regimens (P < 0.001). Furthermore, rifampin combinations with either vancomycin, teicoplanin, fleroxacin, or ciprofloxacin were significantly more bactericidal against adherent bacteria than netilmicin combinations with vancomycin or daptomycin (P < 0.01). Thus, in vivo verification of the pharmacokinetic in vitro model correlated well with the animal model. The in vitro model offers an alternative to ther animal model in experiments that screen and assess antibiotic regimens against device-related infections.