Biases can distort, limit or inhibit the value of mortality data as an epidemiological re source. From 9500 deaths occurring in Naples (Italy during 1994, a random sample of 372 death certificates reporting ill-defined causes and multiple causes of death was extracted. The code for the underlying cause on the death certificate (assigned code) was compared with the cause reattributed with the aid of interview of the certifying physician or clinical records (modified code). The aim was to investigate the extent of misclassification of 'underlying cause' in deaths attributed to ill-defined and/or multiple causes and the shortcomings in the ICD-IX. Ill-defined underlying causes of death (7.0% of death certificates) were cardiovascular diseases, tumours with no specified site or nature, symptoms, signs, ill-defined conditions and senility. There was disagreement between the initially assigned code and the modified code in 53.8% of ill-defined underlying causes; discordance was high for the certificates filled in by the family physician. Multiple causes of death were observed in 23.6% of certificates; of these 59.2% concerned subjects aged 75 years and over at death. Diabetes was always listed in association with other pathologies but neoplasms and traumas were generally listed alone. Disagreement between codes occurred in 48 (54.5%) certificates indicating multiple causes. In 10 of them, death was established as due to a concurrence of causes. As regards ill-defined causes of death, the authors concluded that specific training on certifying procedures would be insufficient on their own; the physician should be made aware that certification is a fundamental requirement for building up epidemiological data. Evidence-based educational interventions are needed. As regards multiple causes of death, multicausal analysis may be indicated for deaths due to a concurrence of causes.