The exposure-response properties of metformin were characterized in 12 subjects with type 2 diabetes mellitus. The time course of drug concentration and effects on fasting plasma glucose and lactic acid concentrations were used from a study in which subjects received 500 mg of metformin twice daily for 5 days followed by 850 mg twice daily for 5 days. Pharmacokinetic sampling included morning trough concentrations obtained on days 7 to 9 and rich sampling (15 time points) on day 10. Fasting plasma glucose and lactic acid concentrations were measured on days 0 to 10 and served as biomarkers of therapeutic effect and tolerability, respectively. A population pharmacokinetic/pharmacodynamic analysis was conducted using nonlinear mixed effects modeling. Metformin pharmacokinetics were described using a 1-compartment model with first-order absorption. Population mean estimates (relative standard error [RSE]) of clearance (CL/F) and volume of distribution were 79.0 L.h(-1) (6.8%) and 648 L (13.8%), respectively. Covariate analyses revealed that creatinine clearance (CL(CR)) significantly influenced metformin CL/F [CL/F = 79.0.(CL(CR)/80)(0.822)]. An indirect response model was applied to describe the antihyperglycemic effect of metformin. Population mean estimates (RSE) of baseline fasting plasma glucose and the drug concentration producing half-maximal effect were 241 mg.dL(-1) (4.6%) and 4.23 mg.L(-1) (31.0%). An empirical linear model was used to describe a slight progressive increase in fasting lactic acid during metformin treatment with an estimated slope coefficient (RSE) of 0.0005 mM.mL.ng(-1) (38.1%). Model evaluation by predictive check and nonparametric bootstrap analysis suggested that the proposed model is robust, and parameter values were estimated with good precision. Simulations suggested that the clinical utility of metformin was maintained over the dose range evaluated with respect to fasting plasma glucose and lactic acid response.