OBJECTIVE: To determine the relationships among serum aminoglycoside clearance, renal aminoglycoside clearance, measured creatinine clearance, and estimated creatinine clearance derived from a standard formula in critically ill patients. SETTING: A ten-bed general ICU in a university hospital. PATIENTS: Eighteen critically ill patients who were being treated with gentamicin or tobramycin for severe infections, and were hemodynamically stable. INTERVENTIONS: The various clearances were measured simultaneously after the administration of a dose of aminoglycoside by assaying serial blood samples for aminoglycoside and creatinine concentration, and by measuring the content of these substances in urine collected over the same time period. OUTCOME MEASURES: The slopes, intercepts and coefficients of determination (r2) of the various regressions were determined, along with the 95% confidence intervals for the prediction of serum aminoglycoside clearance from each other variable. RESULTS: Renal aminoglycoside clearance, creatinine clearance, and estimated creatinine clearance accounted for only 58%, 59%, and 62%, respectively, of the variance in serum aminoglycoside clearance. Only 64% of the variance in renal aminoglycoside clearance was explained by creatinine clearance. Substantial and variable nonrenal aminoglycoside clearance was evident. CONCLUSIONS: The 95% confidence intervals for the prediction of serum aminoglycoside clearance from each index of renal function indicated that none of these indices provided acceptable accuracy for the prediction of serum aminoglycoside clearance and dosage requirements in critically ill patients. Renal aminoglycoside clearance was not better than creatinine clearance in this respect, and thus no other index of renal function is likely to be more accurate. This finding implies that the only accurate method of determining the dose requirements to achieve target serum concentrations in such patients will be individualized pharmacokinetic dosing.