OBJECTIVE To determine whether capillary blood gas measurements provide a clinically acceptable estimate of arterial pH, PCO2, and PO2. DESIGN Prospective convenience sample. SETTING Pediatric intensive care unit at a referral children's hospital. PATIENTS Fifty children > 1 month of age with indwelling arterial catheters. INTERVENTIONS A local anesthetic was applied to the third finger of the hand contralateral to a radial artery catheter. After 90 mins, simultaneous arterial and capillary blood gases were drawn. MEASUREMENTS AND MAIN RESULTS Arterial and capillary pH, PcO2, and PO2 were measured. Heart rate and Wong/Baker faces score were noted before and during capillary blood gas collection to assess discomfort associated with blood collection. There was a strong correlation between capillary and arterial pH (r2 = .903, p < .0001). The relative average bias of the capillary pH was 0.009, with capillary lower than arterial and 95% limits of agreement of +/- 0.032. In all patients, the absolute value of the difference between arterial and capillary pH was < or = 0.05. There was a strong correlation between arterial and capillary PCO2 (r2 = .955, p < .0001). The relative average bias of the capillary PCO2 was 1.6 torr (0.21 kPa), with capillary higher than arterial and 95% limits of agreement of +/- 4.5 torr (+/- 0.6 kPa). In two of 50 patients, the absolute value of the difference between arterial and capillary PCO2 was > 6.5 torr (> 0.87 kPa). Despite a statistically significant correlation between capillary and arterial PO2 (r2 = .358, p < .0001), the absolute value of the difference between arterial and capillary PO2 was > 6.5 torr (> 0.87 kPa) in 42 of 50 patients. Pain, endotracheal intubation, vasoactive drips, or pharmacologic paralysis did not affect accuracy of the capillary pH or PCO2. CONCLUSIONS Capillary blood gases accurately reflect arterial pH and PCO2 in most pediatric intensive care unit patients. Capillary samples did not significantly underestimate arterial hypercarbia or acidosis. This conservative reflection of metabolic status may be particularly useful in hemodynamically stable patients with mild-to-moderate lung disease.