The pharmacodynamics of gemifloxacin against Streptococcus pneumoniae were investigated in a dilutional pharmacodynamic model of infection. Dose fractionation was used to simulate concentrations of gemifloxacin in human serum associated with 640 mg every 48 h (one dose), 320 mg every 24 h (two doses), and 160 mg every 12 h (four doses). Five strains of S. pneumoniae for which MICs were 0.016, 0.06, 0.1, 0.16, and 0.24 mg/liter were used to assess the antibacterial effect of gemifloxacin. An inoculum of 10⁷ to 10⁸ CFU/ml was used, and each experiment was performed at least in triplicate. The pharmacodynamic parameters (area under the concentration-time curve [AUC]/MIC, maximum concentration of drug in serum [C_max]/MIC, and the time that the serum drug concentration remains higher than the MIC [T > MIC]) were related to antibacterial effect as measured by the area under the bacterial-kill curve from 0 to 48 h (AUBKC₄₈) using an inhibitory sigmoid E_max model. Weighted least-squares regression was used to predict the effect of the pharmacodynamic parameters on AUBKC₄₈, and Cox proportional-hazards regression was used to predict the effect of the three pharmacodynamic parameters on the time needed to kill 99.9% of the starting inoculum (T99.9). There was a clear relationship between strain susceptibility and clearance from the model. The simulations (160 mg every 12 h) were associated with slower initial clearance than were the other simulations; in contrast, bacterial regrowth occurred with the 640-mg simulation when MICs were 0.1 mg/liter. The percentage coefficient of variance was 19% for AUBKC₄₈, and the inhibitory sigmoid E_max model best fit the relationship between AUBKC₄₈ and AUC/MIC. C_max/MIC and T > MIC fit less well. The maximum response occurred at an AUC/MIC of >300 to 400. In weighted least-squares regression analysis, there was no evidence that C_max/MIC was predictive of AUBKC₄₈, but both AUC/MIC and T > MIC were. A repeat analysis using only data for which the T > MIC was >75% and for which hence regrowth was minimized indicated that AUC/MIC alone was predictive of AUBKC₄₈. Initial univariate analysis indicated that all three pharmacodynamic parameters were predictive of T99.9, but in the multivariate model only C_max/MIC reached significance. These data indicate that gemifloxacin is an effective antipneumococcal agent and that AUC/MIC is the best predictor of antibacterial effect as measured by AUBKC₄₈. However, C_max/MIC is the best predictor of speed of kill, as measured by T99.9. T > MIC also has a role in determining AUBKC₄₈, especially when the dose spacing is considerable. Once-daily dosing seems most suitable for gemifloxacin.