Acute paraquat self-poisoning is a significant problem in parts of Asia, the Pacific and the Caribbean. Ingestion of large amounts of paraquat results in rapid death, but smaller doses often cause a delayed lung fibrosis that is usually fatal. Anti-neutrophil ('immunosuppressive') treatment has been recommended to prevent lung fibrosis, but there is no consensus on efficacy. To review the evidence for the use of immunosuppression in paraquat poisoning, and to identify validated prognostic systems that would allow the use of data from historical control studies and the future identification of patients who might benefit from immunosuppression. Systematic review. We searched PubMed, Embase and Cochrane databases for 'paraquat' together with 'poisoning' or 'overdose'. We cross-checked references and contacted experts, and searched on [www.google.com] and [www.yahoo.com] using 'paraquat', 'cyclophosphamide', 'methylprednisolone' and 'prognosis'. We found ten clinical studies of immunosuppression in paraquat poisoning. One was a randomized controlled trial (RCT). Seven used historical controls only; the other two were small (n = 1 and n = 4). Mortality in controls and patients varied markedly between studies. Three of the seven non-RCT controlled studies measured plasma paraquat; analysis using Proudfoot's or Hart's nomograms did not suggest that immunosuppression increased survival in these studies. Of 16 prognostic systems for paraquat poisoning, none has been independently validated in a large cohort. The authors of the RCT have performed valuable and difficult research, but their results are hypothesis-forming rather than conclusive; elsewhere, the use of historical controls is problematic. In the absence of a validated prognostic marker, a large RCT of immunosuppression using death as the primary outcome is required. This RCT should also prospectively test and validate the available prognostic methods, so that future patients can be selected for this and other therapies on admission.