BACKGROUND The purpose of this study was to assess early readmission as an indicator of quality of care, to estimate the frequency of avoidable emergency readmission in a French hospital, and then to describe the feasibility and contribution of routine use of this indicator. METHODS A randomly selected sample of 469 readmissions within 30 days after a conventional hospitalization was chosen among the database of 40,242 hospitalizations during the first half of 1997. Two independent practitioners, whose true agreement was measured with a kappa test, studied the features of readmission recorded on the patient files, classing them as "unforeseeable" or "potentially avoidable". Database criteria that could automatically class the readmission in either group were analyzed. RESULTS There were 119 unforeseeable readmissions (25.4%). The two physicians agreed on the unforeseeable nature of 97 of these readmissions and 50 of them were judged avoidable. None of the database criteria allowed identifying all unforeseeable and avoidable readmissions. Readmission via the emergency unit was a sure indicator of unforeseeable readmission in 66% of the cases and of avoidable readmissions in 60%. The frequency of unforeseen readmissions was estimated at 3.9% of all conventional stays during the first half of 1997. The frequency of avoidable readmissions was 1.5%. CONCLUSIONS Unforeseen early readmission can be an indicator of quality of the care taking process. It is however impossible to use the current database to classify with certainty readmissions as "unforeseeable" or "avoidable". Emergency unit readmission could offer a possible approach to measuring the frequency of unforeseen readmission. This ratio can provide caretakers with information concerning the quality of care and thus help in making decisions concerning reorganization for improvement.