Hospital separation data are used to study suicidal behaviour; however, there is little information about the appropriateness of these data for research and planning activities. The study purpose is to examine how consistently hospital separation E-code data reflect suicidal behaviours. Expert clinicians reviewed medical records of individuals who had a separation for self-poisoning to determine whether the self-poisoning was deliberate. Agreement among clinicians was evaluated and latent class analysis performed to derive a summary estimate of the prevalence of deliberate self-poisoning. This estimate was then compared to the prevalence of deliberate self-poisoning based on the external cause of injury (E-codes). Clinicians estimated the prevalence to be 63% higher than the E-code based prevalence. Much larger discrepancies were apparent among older age groups, those whose care was primarily medical in nature and those with a longer length of hospital stay. In acute care settings, self-poisonings among the elderly may not receive adequate attention and/or documentation. Estimating the prevalence of admissions for suicidal behaviour using hospital separation data is of questionable validity, particularly among older age groups.