OBJECTIVE To examine the efficiency of Manitoba hospitals by analysing variations in length of stay for patients with similar characteristics. DESIGN Retrospective study. Multiple regression analyses were used to adjust for patient (case-mix) characteristics and to identify differences in length of stay attributable to the hospital of admission for 14 specific, frequently encountered diagnostic categories and for all acute admissions. SETTING The eight major acute care hospitals in Manitoba. PARTICIPANTS Manitoba residents admitted to any one of the eight hospitals during the fiscal year 1989-90, 1990-91 or 1991-92. Patients transferred to or from another institution, those with atypically long stays and those who died in hospital were excluded. OUTCOME MEASURE Length of hospital stay. RESULTS The length of stay was strongly influenced by hospital of admission, even after adjustment for key patient characteristics. Excluding the most seriously ill patients and those with the longest stays, approximately 186 beds could potentially have been saved if each hospital had discharged its patients as efficiently as the hospital with the shortest overall length of stay. CONCLUSIONS A substantial proportion of days currently invested in treating acute care patients could be eliminated. At least some bed closures in Manitoba hospitals could be accommodated simply through more efficient treatment of patients in the remaining beds, without decreasing access to hospital care.