OBJECTIVE To determine the accuracy and repeatability of pulse dye densitometry (PDD) in measuring blood volume (BV) by comparing it with the conventional method using 51Cr-labeled red blood cells (RI method) and by assessing sequential measurements. METHODS Prospective clinical study. METHODS University hospital. METHODS Eleven adult ICU patients who received cardiac surgery (1st ICU day). METHODS None. RESULTS After injecting indocyanine green (10 or 20 mg) into the right atrium, its arterial concentration was continuously monitored at the nose and finger by PDD, and BV was calculated by back extrapolating the logarithmic dye concentration on the dye elimination curve between 2.5 and 5.5 min after mean transit time to each mean transit time with the least squares method. These measurements were repeated in eight patients and performed only once in the other three, and the BV was measured concurrently by the RI method one time. The Bland-Altman method was used for evaluating differences between methods and within methods. The (percentage) biases and standard deviations between the PDD and RI methods and between the successive measurements by PDD at the finger and nose were 0.26 +/- 0.491 (8.8 +/- 15.3%) and 0.004 +/- 0.251 (0.06 +/- 5.9%) with the probe on a nostril, and 0.16 +/- 0.561 (2.5 +/- 14.4%) and 0.19 +/- 0.311 (4.7 +/- 7.3%) using the finger probe. The bias between methods was less than 10%, and the repeatability of PDD was better. CONCLUSIONS As PDD can measure BV with good repeatability and with a small bias compared to the RI method, serial changes in BV can be evaluated at the bedside of critically ill patients noninvasively and repeatedly.