OBJECTIVE Peripheral capillary oxygen saturation (SpO2) is used as surrogate for arterial blood oxygen saturation. We studied the degree of discrepancy between SpO2 and arterial oxygen (SaO2) and identified parameters that may explain this difference. METHODS We included patients who underwent cardiopulmonary exercise testing at Cleveland Clinic. Pulse oximeters with forehead probes measured SpO2 and arterial blood gas samples provided the SaO2 both at rest and peak exercise. RESULTS We included 751 patients, 54 ± 16 yr old with 53% of female gender. Bland-Altman analysis revealed a bias of 3.8% with limits of agreement of 0.3% to 7.9% between SpO2 and SaO2 at rest. A total of 174 (23%) patients had SpO2 ≥ 5% of SaO2, and these individuals were older, current smokers with lower forced expiratory volume in the first second and higher partial pressure of carbon dioxide and carboxyhemoglobin. At peak exercise (n = 631), 75 (12%) SpO2 values were lower than the SaO2 determinations reflecting difficulties in the SpO2 measurement in some patients. The bias between SpO2 and SaO2 was 2.6% with limits of agreement between -2.9% and 8.1%. Values of SpO2 ≥ 5% of SaO2 (n = 78, 12%) were associated with the significant resting variables plus lower heart rate, oxygen consumption, and oxygen pulse. In multivariate analyses, carboxyhemoglobin remained significantly associated with the difference between SpO2 and SaO2 both at rest and peak exercise. CONCLUSIONS In the present study, pulse oximetry commonly overestimated the SaO2. Increased carboxyhemoglobin levels are independently associated with the difference between SpO2 and SaO2, a finding particularly relevant in smokers.