OBJECTIVES: Most data on self-poisoning in rural Asia have come from secondary hospitals. We aimed to: assess how transfers from primary to secondary hospitals affected estimates of case-fatality ratio (CFR); determine whether there was referral bias according to gender or poison; and estimate the annual incidence of all self-poisoning, and of fatal self-poisoning, in a rural developing-world setting. METHODS: Self-poisoning patients admitted to Anuradhapura General Hospital, Sri Lanka, were reviewed on admission from 1 July to 31 December 2002. We audited medical notes of self-poisoning patients admitted to 17 of the 34 surrounding peripheral hospitals for the same period. FINDINGS: A total of 742 patients were admitted with self-poisoning to the secondary hospital; 81 died (CFR 10.9%). 483 patients were admitted to 17 surrounding peripheral hospitals. Six patients (1.2%) died in peripheral hospitals, 249 were discharged home, and 228 were transferred to the secondary hospital. There was no effect of gender or age on likelihood of transfer; however, patients who had ingested oleander or paraquat were more likely to be transferred than were patients who had taken organophosphorus pesticides or other poisons. Estimated annual incidences of self-poisoning and fatal self-poisoning were 363 and 27 per 100 000 population, respectively, with an overall CFR of 7.4% (95% confidence interval 6.0-9.0). CONCLUSION: Fifty per cent of patients admitted to peripheral hospitals were discharged home, showing that CFRs based on secondary hospital data are inflated. However, while incidence of self-poisoning is similar to that in England, fatal self-poisoning is three times more common in Sri Lanka than fatal self-harm by all methods in England. Population based data are essential for making international comparisons of case fatality and incidence, and for assessing public health interventions.