Most of the knowledge on falls of older persons has been obtained from oral reports that might be biased in many ways. Fall simulations are widely used to gain insight into circumstances of falls, but the results, at least concerning fall detection, are not convincing. Variation of acceleration and maximum jerk of 5 real-world backward falls of 4 older persons (mean age 68.8 years) were compared to the corresponding signals of simulated backward falls by 18 healthy students. Students were instructed to "fall to the back as if you were a frail old person" during experiment 1. In experiment 2, students were instructed not to fall, if possible, when released from a backward lean. Data acquisition was performed using a tri-axial acceleration sensor. In experiment 1, there was significantly more variation within the acceleration signals and maximum jerk was higher in the real-world falls, compared to the fall simulation. Conversely, all values of acceleration and jerk were higher for the fall simulations, compared to real-world falls in experiment 2. The present findings demonstrate differences between real-world falls and fall simulations. If fall simulations are used, their limitations should be noted and the protocol should be adapted to better match real-world falls.