Introduction: Current guidelines and consensus recommend arterial and venous samples as equally acceptable for blood glucose assessment in point-of-care devices, but there is limited evidence to support this recommendation. We evaluated the accuracy of two devices for bedside point-of-care blood glucose measurements using arterial, fingerstick and catheter venous blood samples in ICU patients, and assessed which factors could impair their accuracy. Methods: 145 patients from a 41-bed adult mixed-ICU, in a tertiary care hospital were prospectively enrolled. Fingerstick, central venous (catheter) and arterial blood (indwelling catheter) samples were simultaneously collected, once per patient. Arterial measurements obtained with Precision PCx, and arterial, fingerstick and venous measurements obtained with Accu-chek Advantage II were compared to arterial central lab measurements. Agreement between point-of-care and laboratory measurements were evaluated with Bland-Altman, and multiple linear regression models were used to investigate interference of associated factors. Results: Mean difference between Accu-chek arterial samples versus central lab was 10.7 mg/dL (95% LA -21.3 to 42.7 mg/dL), and between Precision PCx versus central lab was 18.6 mg/dL (95% LA -12.6 to 49.5 mg/dL). Accu-chek fingerstick versus central lab arterial samples presented a similar bias (10.0 mg/dL) but a wider 95% LA (-31.8 to 51.8 mg/dL). Agreement between venous samples with arterial central lab was the poorest (mean bias 15.1 mg/dL; 95% LA -51.7 to 81.9). Hyperglycemia, low hematocrit, and acidosis were associated with larger differences between arterial and venous blood measurements with the two glucometers and central lab. Vasopressor administration was associated with increased error for fingerstick measurements. Conclusions: Sampling from central venous catheters should not be used for glycemic control in ICU patients. In addition, reliability of the two evaluated glucometers was insufficient. Error with Accu-chek Advantage II increases mostly with central venous samples. Hyperglycemia, lower hematocrit, acidosis, and vasopressor administration increase measurement error.