OBJECTIVES: To compare three different methods of falls reporting and examine the characteristics of the data missing from the hospital incident reporting system. DESIGN: Fourteen-month prospective observational study nested within a randomized controlled trial. SETTING: Rehabilitation, stroke, medical, surgical, and orthopedic wards in Perth and Brisbane, Australia. PARTICIPANTS: Fallers (n=153) who were part of a larger trial (1,206 participants, mean age 75.1±11.0). MEASUREMENTS: Three falls events reporting measures: participants' self-report of fall events, fall events reported in participants' case notes, and falls events reported through the hospital reporting systems. RESULTS: The three reporting systems identified 245 falls events in total. Participants' case notes captured 226 (92.2%) falls events, hospital incident reporting systems captured 185 (75.5%) falls events, and participant self-report captured 147 (60.2%) falls events. Falls events were significantly less likely to be recorded in hospital reporting systems when a participant sustained a subsequent fall, (P=.01) or when the fall occurred in the morning shift (P=.01) or afternoon shift (P=.01). CONCLUSION: Falls data missing from hospital incident report systems are not missing completely at random and therefore will introduce bias in some analyses if the factor investigated is related to whether the data is missing. Multimodal approaches to collecting falls data are preferable to relying on a single source alone.