Between January 1987 and December 1991, 68 consecutive patients aged 71.5 +/- 12.0 years underwent percutaneous implantation of a vena caval filter, mainly the LGM (N = 64). Fifty seven patients had pulmonary embolism, 61 had deep vein thrombosis of the lower limbs. The average follow-up interval was 4.9 +/- 3.3 years (7.0 +/- 2.7 years for the patients still alive). The follow-up included a telephonic enquiry to determine the date and cause of death, recurrent deep vein thrombosis and/or pulmonary embolism; surviving patients underwent clinical examination, plain abdominal X-ray with a lateral decubitus view and duplex ultrasonography of the lower limb veins to assess the patency of the filter. Fifty three per cent of the patients died. Four predictive factors for mortality were identified: a contra-indication to anticoagulant therapy, chronic post-embolic cor pulmonale, an indication of prophylactic implantation in the elderly and the presence of underlying malignant disease. There were 5.8% recurrences of pulmonary embolism, 26.1% of lower limb deep vein thrombosis and 25% of filter thrombosis. The only predictive factor of thrombosis was a proximal venous thrombus and was associated in 50% of filter thrombosis. Seventy per cent of the plain abdominal X-rays were abnormal with 9 displacements. 9 migrations and 10 closures of the filters. There was a significant correlation between closure on plain abdominal X-ray and caval thrombosis and between recurrent deep vein thrombosis and caval thrombosis. The frequency of long-term complications after implantation of a caval filter in this study suggests that interruption of the vena cava should be reserved for the only validated indications in the presence of a formal contra-indication to or failure of anticoagulant therapy. Other indications require evaluation with prospective randomised trials.