Background. The influence of dialysis modality on prognosis is controversial. In the absence of randomized trials, epidemiological investigations present the best method for studying the problem. Methods. 4568 haemodialysis (HD) and 2443 peritoneal dialysis (PD) records in 4921 dialysis patients treated between 1990 and 1999 were retrieved from the Danish Terminal Uremia register in order to determine the influence of dialysis form on prognosis. The register is national, comprehensive, and incident. Results. Factors reducing survival included age, cardiovascular disease, malignancy, lung disease, diabetes, alcoholism, haematological disease, but not sex or hypertension. Transplant non-candidacy was associated with an adjusted relative risk of 4.7 (CI 4.0-5.6). PD mortality relative to HD (after correction for comorbidity and transplant candidacy) was 0.65 (CI 0.59-0.72, P < 0.001) on an 'as treated' and 'history' analysis and 0.86 (CI 0.78-0.95, P < 0.01) on an intention-to-treat (ITT) analysis. The difference was confined to the first 2 years of dialysis. Change in dialysis modality was associated with increased mortality, and change from PD to HD with an accelerated mortality for the first 6 months. This was presumably due to the transfer of sick PD patients, but did not explain the difference. The relative advantage of PD was lower for diabetic patients, where it was not significant on ITT analysis. Dialysis prognosis improved by 14% during the period, with similar results for HD and PD patients. PD patients who were subsequently transplanted had a significantly shorter time to onset of graft function (3.5 vs 5.1 days, P < 0.05). Conclusions. These results show a survival advantage for PD during the first 2 years of dialysis treatment. This may be due to unregistered differences in comorbidity at the start of treatment, or may be causal, possibly due to better preservation of residual renal function. The study lends credence to the 'integrative care' approach to uraemia, where patients are started on PD and transferred to HD when PD related mortality increases.