OBJECTIVE: To assess completeness of safety reporting in published clinical trials, including use of incidence, severity, and relationship to drug thresholds for listing of specific adverse events (AEs). METHODS: We used data from previously conducted systematic reviews in three areas: treatment of rheumatoid arthritis (RA) with disease-modifying anti-rheumatic drugs (168 studies, 1969-2001); treatment of migraine with 5HT-1 agonists (38 studies, 1991-2003); and anti-neoplastic treatment of relapsed or refractory non-Hodgkin's lymphoma (NHL) (27 studies, 1991-2003). The type of safety reporting for each study was appraised by two reviewers. RESULTS: Only a minority of studies in each of the clinical areas presented a complete listing of all AEs occuring during the trial (RA 17%, migraine 8%, NHL 30%); a substantial number (10-25%) had no safety data extractable. Among studies with partial AE reporting the thresholds used varied by clinical setting: two-thirds of RA and NHL studies with a reporting threshold used the author- or investigator-attributed relationship to drug to determine which AEs would be listed in published reports, while 71% of migraine studies with a threshold used incidence (e.g. only AEs occuring in more than 5% of patients were listed). The severity threshold (reporting of only serious AEs or only grade 3-4 AEs) was the least common in all three clinical areas examined. No consistent relationship was found between complete AE reporting and study sponsorship (industry vs. non-industry/not reported) or year published (pre vs. post 1995). Smaller studies (<100 patients) were more likely to contain complete AE reporting, perhaps due to the difficulty of providing a comprehensive listing of all events in larger studies. CONCLUSIONS: Incidence and relationship to drug remain common thresholds for AE reporting in published clinical trials. Early detection of rare or unanticipated events by meta-analysis of published trial data is thus made more challenging.