This study investigated variations in the prediction of blood volumes from a single measurement of red cell volume (RCV) with (51)Cr or plasma volume (PV) with (125)I human serum albumin (HSA). In 111 subjects, fractional and whole blood volumes were estimated from separate direct measurements of RCV and PV. The f ratio (body to venous hematocrit) was also determined. There was a very good correlation between (125)I-HSA measured PV (2857 +/- 822 ml) and that estimated with (51)Cr-tagged red blood cells (2864 +/- 747 ml) (r = 0.936, p = 0.000) and also between (51)Cr measured RCV (2600 +/- 774 ml) and that estimated with (125)I-HSA (2589 +/- 843 ml) (r = 0.944, p = 0.000). The 95% limits of agreement (mean +/- 2SD of differences, relative to the mean of paired data) ranged -0.2% +/- 20.3% and 0.4% +/- 21.4%, respectively. The 95% prediction intervals of measured from estimated fractional blood volumes spanned +/-20.3% and +/-19.5%, respectively, relative to the predicted values with regression equations. Proportional degrees of inaccuracy were found in whole blood volume estimations. The f ratio was inconstant and correlated with PV and the body hematocrit. We conclude that blood volumes can be determined reliabily only with direct measurements of RCV and PV. Estimated blood volumes may lead to misconceptions.