In order to determine whether measurement of arterial oxygen saturation (SaO2) could identify patients with obstructive sleep apnea (OSA), 98 consecutive patients referred for assessment of snoring and/or daytime somnolence were assessed clinically and then underwent both unsupervised oximetry in their homes and formal polysomnography. Clinical assessment identified patients with an apnea+hypopnea index (AHI) > or = 15 events per hour with a sensitivity of 79% and a specificity of 50%. Home oximetry analyzed by counting the number of arterial oxygen desaturations recorded was inferior to clinical assessment. For desaturations of 2% or more from baseline, desaturation index (DI) > or = 15 per hour identified patients with AHI > or = 15 with sensitivity 65% and specificity 74%; for 3% desaturations, sensitivity was 51% and specificity 90%; and for 4% desaturations, sensitivity was 40% and specificity 98%. From the oximetry data, the percentage of time spent at SaO2 below 90% (CT90) was also calculated. CT90 > or = 1% identified patients with AHI > or = 15 with sensitivity 93% and specificity 51%; for patients with AHI > or = 15 ultimately given nasal continuous positive airway pressure (CPAP), the sensitivity of a CT90 > or = 1% was 100%. We concluded that home oximetry with CT90 < 1% practically excludes clinically significant OSA. Conversely, home oximetry with DI > or = 15 for 4% desaturations makes OSA likely: the positive predictive value for OSA is 83% if the pretest probability of OSA is 30% and over 90% if the pretest probability is at least 50%.