BACKGROUND: One of the key mechanisms underlying OSA is reduced pharyngeal muscle tone during sleep. Data suggest that pharmacologic augmentation of central serotonergic/ adrenergic tone increases pharyngeal muscle tone. RESEARCH QUESTION: We hypothesized that venlafaxine, a serotonin-norepinephrine reuptake inhibitor, would improve OSA severity. STUDY DESIGN AND METHODS: In this mechanistic, randomized, double-blind, placebo-controlled crossover trial, 20 patients with OSA underwent two overnight polysomnograms >= 4 days apart, receiving either 50 mg of immediate-release venlafaxine or placebo before bedtime. Primary outcomes were the apnea-hypopnea index (AHI) and peripheral oxygen saturation (SpO2) nadir, and secondary outcomes included sleep parameters and pathophysiologic traits with a view toward understanding the impact of venlafaxine on mechanisms underlying OSA. RESULTS: Overall, there was no significant difference between venlafaxine and placebo regarding AHI (mean reduction, -5.6 events/h [95% CI, -12.0 to 0.9]; P = .09) or SpO(2) nadir (median increase, +1.0% [-0.5 to 5]; P = .11). Venlafaxine reduced total sleep time, sleep efficiency, and rapid eye movement (REM) sleep, while increasing non-REM stage 1 sleep (P-all < .05). On the basis of exploratory post hoc analyses venlafaxine decreased ("improved") the ventilatory response to arousal (-30%; P = .049) and lowered ("worsened") the predicted arousal threshold (-13%; [P = .02]; ie, more arousable), with no effects on other pathophysiologic traits (P-all >= .3). Post hoc analyses further suggested effect modification by arousal threshold (P = .002): AHI improved by 19% in patients with a high arousal threshold (-10.9 events/h [-3.9 to -17.9]) but tended to increase in patients with a low arousal threshold (+7 events/h [-2.0 to 16]). Other predictors of response were elevated AHI and less collapsible upper airway anatomy at baseline (vertical bar r vertical bar > 0.5, P <= .02). INTERPRETATION: In unselected patients, venlafaxine simultaneously worsened and improved various pathophysiologic traits, resulting in a zero net effect. Careful patient selection based on pathophysiologic traits, or combination therapy with drugs countering its alerting effects, may produce a more robust response.