Objective To evaluate the accuracy of five blood glucose measurements commonly used in ICU and to determine the potential factors interfering the accuracy. Method This prospective study carried out in consecutively enrolled 49 patients stayed more than 48 hours in the medical ICU of Peking Union Medical College Hospital from November 2007 to January 2008. A total of 130 blood samples were measured to determine blood glucose with five different methods, and the biochemistry analyzer in central laboratory was regarded as reference method for comparison with other four methods, ( 1 ) capillary blood/glucometer;(2) arterial blood/glucometer; (3) arterial blood/blood gas analyzer; and (4) arterial blood/biochemistry analyzer. The accuracy of a method tested was judged by the difference in level of blood glucose between it and reference method, correlation coefficient, bias correction factor and discrepancy. The independent factors for the discrepancies were identified by multivariate logistic regression. Results The values of differences in level of blood glucose between above four methods and the reference were ( 1.7 ± 1.4) mmol/L,( 1.6 ± 1.4 ) mmol/L, ( 1.1 ± 1.2) mmol/L, and (0.5 ± 1.2 ) mmol/L, respectively. The rates of discrepancy were 66.9%, 63.8%, 40.0% and 23.8%. The correlation coefficients were 0. 844, 0. 845, 0. 895and 0. 896. The bias correction factors were 0. 821,0.831,0.914 and 0. 979. Decrease in hematocrit was statistically associated with the discrepancy between glucometer analysis methods and the reference, with OR of 0.923 for capillary blood ( P = 0.03 ) and 0. 912 for arterial blood( P = 0.014), respectively. Conclusions Glucometer analysis poorly correlated with reference method and blood gas analysis in ICU patients. Decrease in hematocrit significantly increased the degree of discrepancy in glucose measurements between glucometer analysis and the reference.