The function of 221 dialysis shunts was investigated retrospectively in 111 patients of an outpatient haemodialysis centre. The mean age of the patients was 58 years. Of 221 recorded shunts, 171 were direct arteriovenous fistulas, while 50 were interposition grafts. The longest observation period was almost 20 years (227 months). The cumulative function rate was evaluated by the life table method; primary patency (until first revision) was differentiated from total function (revisions included). The direct fistulas as compared to the grafts had a significantly (P less than 0.05) higher primary function rate (median 35 vs. 13 months) and after 36 months also a significantly longer total function (median 76 vs. 28 months). Complications were significantly (P less than 0.05) more frequent in grafts than in fistulas (61/50 vs 120/171). Surgical revisions of fistulas failed significantly more often (P less than 0.05) than revisions of grafts (137/171 versus 26/50). Grafts on the forearm had significantly worse results as compared to grafts or fistulas on the upper arm (P = 0.03). In contrast to age, sex and underlying disease of patients a blood pressure less than 130/70 mmHg was a significant risk factor for shunt failure (P = 0.002). Analysis of primary function shows (clearer than the total patency rate) that a direct fistula is a better vascular access in haemodialysis patients than a graft.