The aims of this study were to audit the monitoring of aminoglycoside treatment to develop a method for measurement of the cost and outcome of treatment, and to assess the accuracy of a previously published sepsis score to predict cost and outcome. Measurements of costs and outcomes were also made for all patients (n = 85) with Gram-negative bacteraemia. Of these, 40 received an aminoglycoside and an additional 215 patients received aminoglycosides for other indications (total aminoglycoside patients 255; total patients 300). There were no interpretable assays for 82 (32%) of the aminoglycoside patients and only 33/173 (19%) patients assayed had first peak serum concentrations within the recommended range of 8-10 mg/L. Median costs of aminoglycoside treatment were 599 pounds in neutropenic patients, 471 pounds in ICU patients and 185 pounds in other patients. In the bacteraemic patients, median costs of aminoglycoside regimens (278 pounds) were higher than for non-aminoglycoside regimens (97 pounds). Death in hospital was twice as common in bacteraemic patients (20% versus 10%) and there was a stepwise increase in rate of mortality with sepsis scores. Treatment costs were markedly higher in patients who failed to respond to initial treatment, the mean difference in cost was 418 pounds per patient (95% CI 89 pounds - 747). Sepsis scores only explained 2.6% of the variance in treatment costs, and 22 patients with zero sepsis scores received prolonged courses of i.v. antibiotic treatment at an average cost of 209 pounds per patient. In conclusion, aminoglycoside regimens rarely conformed to accepted standards of care and treatment failure was associated with markedly increased treatment costs. Three readily measurable indicators of adverse outcome were identified (death in hospital, change of i.v. treatment and readmission within 2 weeks of discharge) and all were related to initial severity of illness as measured by sepsis score. The sepsis score may prove useful for assessment of individual risk but would benefit from further analysis to validate and possibly reduce the number of items in the score.