Compact analyzers suited to near-patient testing estimate hematocrit by measuring the conductivity of undiluted blood. We evaluated the accuracy of hematocrit determination of one such analyzer (Instrumentation Laboratory BGE analyzer) against an automated cell counter (EPC) and packed cell volume (PCV) microhematocrit. When specimens (n = 34) from outpatient and ward patients were analyzed with all three methods, the BGE analyzer correlated well with both EPC and PCV hematocrit determinations (BGE = 1.00 PCV + 0.3%, S(y)/x = 2.0%), suggesting that all three methods are similar in performance for most patients. However, a patient with increased plasma osmolality showed significant decreases in BGE and PCV hematocrits relative to the EPC method. The differences in hematocrit measurements could be reproduced by adding solutes to blood in vitro or by modifying the plasma osmolality of rats in vivo. Samples from patients undergoing cardiac surgery, whose blood had large changes in protein concentration, showed discrepancies between hematocrits by conductivity and other methods; similar effects could be produced by changes in protein concentration or in vitro addition of polyethylene glycol. We conclude that conductivity measurements provide accurate hematocrit results for physiologically normal subjects but not for some intensive-care and surgical patients.