Life expectancy among individuals with spinal cord injuries (SCI) has remained lower than in the normal population, even with optimal medical management. But signi®cant improvement has been achieved, as will be illustrated in this retrospective study of an unselected group of traumatic survivors of SCI, dead or still living. There has been a complete follow-up over 4 decades, information being obtained from available medical records, death certi®cates, and post mortem records. The survey included a total of 888 individuals (713 men and 175 women) who had survived the injury and primary treatment and were rehabilitated at the centre for Spinal Cord Injured in Hornbñk, Denmark. At the end of the follow-up, 31st December 1992, 236 (197 men and 39 women) had died. The commonest causes of death were lung diseases, particularly pneumonia; suicide; and ischaemic heart disease. Among functionally complete tetraplegic individuals there was a recognizably high percentage of deaths from pneumonia, and among the least disabled individuals (Frankel class E) we found a high frequency of suicides. The Standardised Mortality Ratios (SMRs) were highest for septicaemia, followed by uraemia and pneumonia. A signi®cant decrease in the overall mortality was observed from the ®rst (1953 ± 1973) to the second half of the observation period (1972 ± 1992). Similarly the survival curves for both men and women demonstrate that the gap in survival probability between the normal population and the SCI has diminished considerably from the early to the later period. Likewise, except for suicide and ischaemic heart disease, a decrease in SMRs was seen for all causes of death. In particular there were large decreases related to lung embolus, septicaemia, pneumonia, and uraemia. The patterns of causes of death in the study group begin to approximate those of the general population, though many cause-speci®c deaths for SCI remain substantially above the normal population. Continuous improvement in preventive measures as well as treatment procedures is still necessary.