The pharmacokinetics and pharmacodynamics of bacterially synthesized granulocyte-macrophage colony stimulating factor (GM-CSF) were studied in 33 patients. GM-CSF (0.3-30 micrograms/kg/d) was administered subcutaneously (s.c.) or intravenously (i.v.) by bolus or 2 h infusion for 10 d to three patients at each dose level, and serum levels measured by enzyme-linked immunosorbent assay (ELISA). Relationships between pharmacokinetic variables and GM-CSF efficacy and toxicity were investigated. GM-CSF appeared more effective s.c. than i.v. Correlations with response improved when patients with neutropenia or marrow infiltration by tumour were excluded. For the remaining patients, the correlation coefficients between the day 10 leucocyte count was highest with dose (R = 0.629, P less than 0.01), the duration GM-CSF concentrations exceeded 1 ng/ml (r = 0.524, P less than 0.05) and area under the concentration-time curve (R = 0.508, P less than 0.05). There was no correlation with peak serum GM-CSF (Cmax) (R = 0.188, P = NS). In contrast, the first dose reaction (characterized by hypoxaemia and hypotension) was associated with high Cmax (P less than 0.01) and i.v. administration. Fever, liver enzyme elevation and pericarditis occurred at 10-30 micrograms/kg and were not influenced by route. Since the variables correlating with response differed from those correlating with these toxicities, it appears that the therapeutic index for GM-CSF can be increased by modifying the method of administration.