The charts of all patients with diving-related accidents presenting between 1983 and 1991 were reviewed. Individuals who sustained a neurologic deficit attributable to occlusion of part of the cerebral circulation within 10 minutes of surfacing from a dive and who had a depth time profile less than 80% of the U.S. Navy "no-stop" limits were considered to have had an arterial gas embolism. Their records were reviewed to determine the hematocrit upon first presentation and the final hematocrit prior to discharge or death. Twenty-three patients had hematocrit determinations on more than one occasion. There was a significant decrease from initial to final hematocrit for these patients. No source of significant blood loss was identified in any patient. A significant correlation was found between the magnitude of the fall in hematocrit and the eventual neurologic outcome. We conclude that gas embolism diagnosed using the above criteria produces hemoconcentration, and the degree of hemoconcentration correlates with the severity of the gas embolism. These observations suggest that the pathophysiology of gas embolism is more complex than previously thought and must include diffuse endothelial injury resulting in leak of fluid from the intravascular space.