OBJECTIVES: the main objective was to evaluate if the admission functional independence measure (FIM) score could be used to predict the risk of falls in geriatric inpatients. DESIGN: a 10-year retrospective study was performed. SETTING: the study was conducted in a 298-bed geriatric teaching hospital in Geneva, Switzerland. SUBJECTS: all patients discharged from the hospital from 1 January 1997 to 31 December 2006 were selected. MAIN OUTCOME MEASURES: measures used were FIM scores at admission using the FIM instrument and number of falls extracted from the institution's fall report forms. RESULTS: during the study period, there were 23,966 hospital stays. A total of 8,254 falls occurred. Of these, 7,995 falls were linked to 4,651 stays. Falls were recorded in 19.4% of hospital stays, with a mean incidence of 7.84 falls per 1,000 patients-days. Although there was a statistically significant relationship between total FIM score, its subscales, and the risk of falling, the sensitivity, specificity, positive predictive value and negative predictive value obtained with receiver operating characteristic curves were insufficient to permit fall prediction. This might be due in part to a non-linear relationship between FIM score and fall risk. CONCLUSION: in this study, the FIM instrument was found to be unable to predict risk of falls in general geriatric wards.