The measurement of peripheral oxygen saturation (SpO2) in neonatal intensive care units (NICUs) poses a significant challenge. Motion artifacts due to the patient's limb motion induce many false alarms, which in turn cause an additional workload for the medical staff and anxiety for the parents. We developed a reflectance pulse oximeter dedicated to be placed at the patient's forehead, which is less prone to such artifacts. We trained our algorithms for SpO2 estimation on 8 adult healthy volunteers participating in a controlled desaturation study. We then validated our SpO2 monitoring system on 25 newborn patients monitored in an NICU. We further evaluated the versatility and resilience to low signal-tonoise ratios (SNR) of our solution by testing it on signals acquired in a low-perfusion region (upper right part of the chest) of our adult volunteers. We obtained an SpO2 estimation accuracy ($A _{\mathbf {rms}}$) of 1.9 % and 3.1 % at the forehead and the chest in our adult volunteers, respectively. These performances were obtained after automatic rejection of 0.1 % and 30.0 %, respectively, of low-SNR signals by our dedicated quality index. In the dataset recorded on newborn patients in the NICU, we obtained an accuracy of 3.9 % after automatic rejection of 11.7 % of low-SNR signals by our quality index. These analyses were carried out following the procedures suggested by the ISO 80601-2-61:2011 standard, which specifies a target $A _{\mathbf {rms}} \le $ 4 % for SpO2 monitoring applications. These promising results suggest that reflectance pulse oximeters can achieve clinically acceptable accuracy, while being placed at locations less sensitive to limb motion artifacts - such as the forehead - thereby reducing the amount of SpO2-related false alarms in NICUs.