OBJECTIVES--To determine the accuracy of (1) hospital discharge data and (2) death certificates, coded as motor neuron disease (MND). DESIGN--Comparison of data from The Scottish Motor Neuron Disease Register (SMNDR) with routinely collected Scottish Hospital In-Patient Statistics (SHIPS) and death certificate coding. SETTING--Scotland UK. PATIENTS--1) 379 adults (> 15 years) discharged for the first time from a Scottish hospital in 1989-90 and (2) 281 deaths in the same period assigned to the International Classification of Diseases (ICD)-9, category 335 (MND). MAIN OUTCOME MEASURES--The sensitivity and positive predictive value of a diagnosis of MND as retrieved by (1) the Information and Statistics Division of the Common Services Agency for the Scottish Health Service for morbidity data and (2) the Registrar General's office for mortality data, using the SMNDR as the 'gold standard'. RESULTS--(1) Thirty per cent of adult patients identified as having MND by SHIPS did not have this disease and 23% of patients with MND did not appear on SHIPS. The sensitivity of a diagnosis of MND, as retrieved by SHIPS, was 84% and the positive predictive value was 70% overall. Miscoding of patients with pseudobulbar palsy caused by cerebrovascular disease was the major source of false positive error. The incidence of adult onset sporadic MND was over estimated by SHIPS by a factor of 1.6. (2) Mortality data were more accurate, with a false negative rate of 6% and a positive predictive value of 90%. CONCLUSIONS--Coded hospital discharge data are an inaccurate record of a diagnosis of MND and cannot, in their present form, be used as a reliable measure of disease incidence in Scotland. Greater care is required in the preparation of discharge summaries and coding if these data are to be useful for health care planning and epidemiological research. SHIPS is, however, an important source of information to achieve a complete sample of patients with MND. There is also a problematic false positive rate for mortality data but this source more closely approximates true incidence.