Objective To explore the potential mechanisms underlying the prominent efficiency of hyperbaric oxygen therapy (HBOT) in the treatment of severe coronavirus disease 2019 (COVID-19) patients. Methods Five COVID-19 patients, aged from 24 to 69 years old, received HBOT after routine therapies failed to stop the deterioration and progressive hypoxemia in General Hospital of the Yangtze River Shipping. The procedure of HBOT was as follows:compressed to 2.0 ATA (0.1 MPa gauge pressure, patient 1) or 1.6 ATA (0.06 MPa gauge pressure, patient 2-5) at a constant rate for 15 min, maintained for 90 min (first treatment) or 60 min (subsequent treatment), then decompressed to normal pressure for 20 min, once a day; the patients inhaled oxygen with the mask of Built-in-Breathing System continuously; and HBOT was ended when the daily mean pulse oxygen saturation (SpO₂) in wards was above 95% for two days. The symptoms, respiratory rate (RR), SpO₂, arterial blood gas analysis, blood routine, coagulation function, high-sensitivity C-reactive protein (hs-CRP) and chest computed tomography (CT) were collected. Paired t test was used to compare each index before and after treatment. Results After the first HBOT, the symptoms and signs of the five patients began to improve. Supine breathlessness disappeared after HBOT for four times, and digestive tract symptoms completely disappeared and only mild chest pain and breathlessness at rest and in motion remained after HBOT for five times. After finishing HBOT, the RR of the patients was significanlty lower than that before HBOT ([20.80±2.28] min^-1 vs[27.20±5.40] min^-1, P<0.05). After finishing HBOT, daily SpO₂ in wards was increased day by day, and the daily mean SpO₂ recovered to more than 95% after the first, second, third, third and sixth HBOT in the five patients, respectively. After the first HBOT decompression, SpO₂ was (93.60±0.07)%, which was signficantly higher than that before HBOT ([73.20±6.42]%) (P<0.05). SpO₂ values before compression of the second and third HBOT were signficantly higher than that before the first HBOT (both P<0.05). There was no significant difference in the SpO₂ immediately before and after the third HBOT (P>0.05). Before HBOT, the arterial partial pressure of carbon dioxide (PaCO₂) of the patients was (31.48±3.40) mmHg (1 mmHg=0.133 kPa), which was lower than the normal range (35-45 mmHg). After finishing HBOT, arterial partial pressure of oxygen ([130.20±18.58] mmHg), arterial oxygen saturation ([98.40±0.55]%), lymphocyte proportion (0.207 8±0.074 2) and lymphocyte count ([1.09±0.24]×10⁹/L) were significantly higher than those before HBOT ([61.60±15.24] mmHg,[73.20±6.43]%, 0.094 6±0.062 1, and[0.61±0.35]×10⁹/L), while the levels of fibrinogen ([2.97±0.27] g/L) and hs-CRP ([7.76±6.95] mg/L) were significantly lower than those before HBOT ([4.45±0.94] g/L and[30.36±1.27] mg/L) (all P<0.05). The levels of lacttic acid and D-dimer were decreased after HBOT versus before HBOT ([1.13±0.10] mmol/L vs[2.16±1.71] mmol/L,[0.42±0.13] mg/L vs[1.84±1.29] mg/L), but the differences were not significant (both P>0.05). All the five patients had typical lung CT imaging changes of severe COVID-19 before HBOT, which were improved after HBOT. Conclusion Systemic hypoxia induced by persistent hypoxemia may be the main reason for the deterioration of severe COVID-19. The respiratory dysfunction of COVID-19 is mainly alveolar gas exchange dysfunction. HBOT may be the best way to correct the progressive hypoxemia which can not be controlled by atmospheric oxygen supply in severe COVID-19 patients. HBOT can provide enough oxygen supply for the continuous hypoxia tissues, and is beneficial to the recovery of immune function, circulatory function and stress level, so as to improve the condition of patients.