A computer-based system for storing items of medical record data including a complete coded list of diagnoses was established in 1971 for all patients admitted to a research-oriented medical ward. An analysis was made to assess the usefulness of such an electronically-stored data base from the stand-point of Unit activity, disease correlations, stability of diagnostic criteria, completeness of diagnosis lists and accuracy of coding. Over six years, 1972-77, there were 3569 admissions of which 1679 were first admissions. The most frequently made principal diagnoses were cerebrovascular accident (8%), myocardial infarction (5.4%), ischaemic heart disease (2.9%), rheumatoid arthritis (2%), duodenal ulcer (2%) and systemic lupus erythematosus (1.9%); the 33 most frequent principal diagnoses accounted for only 45% of all principal diagnoses made. The duration of stay (mean 16 days) was shown to depend on principal diagnosis but not significantly on age. Variability in annual incidence was significant for 24% of diagnoses; for some diagnoses this was readily explainable by extraneous causes, but for others it suggested an "instability" of criteria for that diagnosis. The probability of a minor diagnosis being "overlooked", using Dupuytren's contracture as an example, was shown to be high (50%). The rate of miscoding a diagnosis was 5%. This study, despite its illustration of the "softness' of diagnosis making in routine hospital ward practice, illustrates the potential or processing hospital diagnosis data by computer.