We attempted to predict revolving door (RD) patterns of admission (four or more admissions with less than 2.5 years between consecutive admissions) in a random sample of 10% of all first admissions to psychiatric hospitals and psychiatric wards of general hospitals in Israel from 1983 to 1990 with follow-up into 1993. This included 4570 hospitalizations of 2220 patients. Data were extracted from the National Psychiatric Case Registry of the Ministry of Health. Almost 59% of the sample had only one admission, 41% had two or more, 23% had three or more, and 14% had four or more admissions. Patients with four or more admissions were all RD patients. They had an average of 200 days between admissions. The average number of admissions for RD patients was 6.17, and the average number of years between the first admission and the last admission was 3.28 years. Using discriminant analysis were correctly predicted 73.9% of the non-RD group (about chance level since 80% of the cases were non-RD) and 71.2% of the RD group (considerably better than chance, only 12.0% of the sample were RD). The main predictors of RD in descending order were not being married at the time of first hospitalization, unemployment and more severe initial diagnosis. The minor predictors were older age, more education and longer first admission. Substance abuse, patients ability to care for their affairs, voluntary status of first admission and suicide attempts did not predict RD. The predictors of RD were almost the same as predictors of more than one admission. We were not able to identify a variable that clearly differentiated between the two or more, three or more and four or more admissions groups. Variability between hospitals is also presented.