The 'Hemocue' device for rapid estimation of haemoglobin concentration was evaluated in a clinical setting. Repeatable accuracy of capillary, venous and arterial samples was examined and then compared with standard laboratory venous haemoglobin estimates using a 'Coulter JT' analyser in 42 patients. The mean values for haemoglobin (g/l) and coefficient of variation were capillary 108.2 (8.0); venous 104.9 (2.2); arterial 105.9 (2.0); and laboratory venous 104.6 (1.3). Although the mean haemoglobin values were similar, capillary samples were significantly less repeatable than venous or arterial samples (Pitman test, P < 0.001). Comparison of variance between the laboratory sample and each sampling technique demonstrated that capillary samples were significantly more variable than venous or arterial samples. Peripheral skin temperature did not influence the accuracy of capillary samples. Hemocue estimations of venous samples were found to be as accurate as laboratory estimations. The lack of repeatable accuracy of capillary estimations was sufficiently large that their use cannot be recommended in clinical practice.