SummaryOne night of a sleep study is the standard for diagnosis and exclusion of obstructive sleep apnea. Single testing requires high sensitivity of the test method and a stable disease of interest to warrant a low rate of false‐negative tests. Obstructive sleep apnea is diagnosed and graded by conventional thresholds of apneas and hypopneas per hour of sleep, and treatment is usually initiated in the presence of symptoms. The aim of this study was to assess night‐to‐night variability of obstructive sleep apnea to reassess the current practice. Seventy‐seven patients previously diagnosed with obstructive sleep apnea, randomised to continuous positive airway pressure withdrawal within four trials, performed nightly pulse‐oximetry over 2 weeks while off continuous positive airway pressure. The main outcome of interest was the coefficient of variation of the oxygen desaturation index marking night‐to‐night variability in obstructive sleep apnea. Obstructive sleep apnea was categorised according to conventional thresholds using oxygen desaturation index (no obstructive sleep apnea: <5 per h; mild: 5–15 per h; moderate: 15–30 per h; and severe: >30 per h). High night‐to‐night variability of obstructive sleep apnea was evidenced by a coefficient of variation of oxygen desaturation index of 31.1% (SD 16.5). Differences in oxygen desaturation index of >10 per h between nights were found in 84.4% and shifts in obstructive sleep apnea severity category in 77.9% of patients. The probability of missing moderate obstructive sleep apnea was up to 60%. Variability was higher in less severe obstructive sleep apnea. Obstructive sleep apnea shows a considerable night‐to‐night variability. Single‐night diagnostic sleep studies are prone to miscategorise obstructive sleep apnea if arbitrary thresholds are used. Thus, treatment decisions should be based less on the conventional derivatives from sleep studies, especially in patients with less severe obstructive sleep apnea. Clinical trial registration: www.controlled-trials.com (ISRCTN 93153804, ISRCTN 73047833) and www.clinicaltrials.gov (NCT01332175 & NCT02050425).