To examine the reliability of adverse incident-reporting systems we carried out a retrospective review of the mother and baby case notes from a series of 250 deliveries in each of two London obstetric units. Notes were screened for the presence of adverse incidents defined by lists of incidents to be reported in accordance with unit protocols. We assessed the percentage of adverse incidents reported by staff to the maternity risk manager at each unit; the percentage of incidents detected by each risk manager, but not reported; and the percentage of incidents identified only by retrospective case note review. A total of 196 adverse incidents was identified from the 500 deliveries. Staff reported 23% of these and the risk managers identified a further 22%. The remaining 55% of incidents were identified only by retrospective case-note review and not known to the risk manager. Staff reported about half the serious incidents (48%), but comparatively few of the moderately serious (24%) or minor ones (15%). The risk managers identified an additional 16% of serious incidents that staff did not report. Drug errors were analysed separately; only two were known to the risk managers and a further 44 were found by case-note review. Incident-reporting systems may produce much potentially valuable information, but seriously underestimate the true level of reportable incidents. Where one risk manager covers an entire trust, rather than a single unit, reporting rates are likely to be very much lower than in the present study. Greater clarity is needed regarding the definition of reportable incidents (including drug errors). Staff should receive continuing education about the purposes and aims of clinical risk management and incident reporting and consideration should be given to designating specific members of staff with responsibility for reporting.