OBJECTIVE: To increase understanding of decisions about inpatient admission, a four-step algorithm was used to examine 2,073 consecutive visits to a public hospital psychiatric emergency room, 684 of which resulted in admission. METHODS: Admission decision outcomes and patient data were cross-tabulated to identify conditions, or rules, under which outcome was almost certain. Discriminant function models were then made of individual clinicians' decision-making process and of individual diagnostic groups. To understand cases not covered in previous steps, a third discriminant function model was constructed. RESULTS: The four- step method successfully predicted outcomes in 85 percent of cases at a minimum of an 80 percent confidence level. The variables of psychosis and violence combined into the most powerful predictor of admission. Twelve rules that applied to 41.4 percent of all cases were found. Eleven models of individual clinicians' decision policies applied to slightly more half of all cases and successfully classified about 95 percent of them. Eleven models of diagnostic groups applied to 93.2 percent of all cases and correctly predicted about 75 percent. The final discriminant model for the 171 cases not covered by the by the first three steps correctly classified about 90 percent of residual cass. CONCLUSIONS: Psychiatric admission decisions are influenced by multiple variables that should be studied by examining general admission criteria and differences between clinicians.