<h3>Importance</h3> The effect of high-flow oxygen therapy vs conventional oxygen therapy has not been established in the setting of severe COVID-19. <h3>Objective</h3> To determine the effect of high-flow oxygen therapy through a nasal cannula compared with conventional oxygen therapy on need for endotracheal intubation and clinical recovery in severe COVID-19. <h3>Design, Setting, and Participants</h3> Randomized, open-label clinical trial conducted in emergency and intensive care units in 3 hospitals in Colombia. A total of 220 adults with respiratory distress and a ratio of partial pressure of arterial oxygen to fraction of inspired oxygen of less than 200 due to COVID-19 were randomized from August 2020 to January 2021, with last follow-up on February 10, 2021. <h3>Interventions</h3> Patients were randomly assigned to receive high-flow oxygen through a nasal cannula (n = 109) or conventional oxygen therapy (n = 111). <h3>Main Outcomes and Measures</h3> The co–primary outcomes were need for intubation and time to clinical recovery until day 28 as assessed by a 7-category ordinal scale (range, 1-7, with higher scores indicating a worse condition). Effects of treatments were calculated with a Cox proportional hazards model adjusted for hypoxemia severity, age, and comorbidities. <h3>Results</h3> Among 220 randomized patients, 199 were included in the analysis (median age, 60 years; n = 65 women [32.7%]). Intubation occurred in 34 (34.3%) randomized to high-flow oxygen therapy and in 51 (51.0%) randomized to conventional oxygen therapy (hazard ratio, 0.62; 95% CI, 0.39-0.96;<i>P</i> = .03). The median time to clinical recovery within 28 days was 11 (IQR, 9-14) days in patients randomized to high-flow oxygen therapy vs 14 (IQR, 11-19) days in those randomized to conventional oxygen therapy (hazard ratio, 1.39; 95% CI, 1.00-1.92;<i>P</i> = .047). Suspected bacterial pneumonia occurred in 13 patients (13.1%) randomized to high-flow oxygen and in 17 (17.0%) of those randomized to conventional oxygen therapy, while bacteremia was detected in 7 (7.1%) vs 11 (11.0%), respectively. <h3>Conclusions and Relevance</h3> Among patients with severe COVID-19, use of high-flow oxygen through a nasal cannula significantly decreased need for mechanical ventilation support and time to clinical recovery compared with conventional low-flow oxygen therapy. <h3>Trial Registration</h3> ClinicalTrials.gov Identifier:NCT04609462