There is dispute as to whether paradoxical gas embolism is an important aetiological factor in neurological decompression illness, particularly when the spinal cord is affected. We performed a blind case-controlled study to determine the relationship between manifestations of neurological decompression illness and causes in 100 consecutive divers with neurological decompression illness and 123 unaffected historical control divers. The clinical effects of neurological decompression illness (including the sites of lesions and latency of onset) were correlated with the presence of right-to-left shunts, lung disease and a provocative dive profile. The prevalence and size of shunts determined by contrast echocardiography were compared in affected divers and controls. Right-to-left shunts, particularly those which were large and present without a Valsalva manoeuvre, were significantly more common in divers who had neurological decompression illness than in controls (P<0.001). Shunts graded as large or medium in size were present in 52% of affected divers and 12.2% of controls (P<0.001). Spinal decompression illness occurred in 26 out of 52 divers with large or medium shunts and in 12 out of 48 without (P<0.02). The distribution of latencies of symptoms differed markedly in the 52 divers with a large or medium shunt and in the 30 divers who had lung disease or a provocative dive profile. In most cases of neurological decompression illness the cause can be determined by taking a history of the dive profile and latency of onset, and by performing investigations to detect a right-to-left shunt and lung disease. Using this information it is possible to advise divers on the risk of returning to diving and on ways of reducing the risk if diving is resumed. Most cases of spinal decompression illness are associated with a right-to-left shunt.