A　prediction　method　of　elderly　fall　based　on　RBF　neural　network　and　multi-sensor　information　fusion　was　proposed　in　this　paper,　which　could　help　the　user　of　elderly-assistant　robot　walk　outside　safe　and　reduce　the　damage　caused　by　falls.　Firstly,　overall　scheme　of　the　elderly　fall　prediction　system　was　designed.　Tactile　sensors,　tri-axial　acceleration　sensor　and　gyroscope　were　used　to　collect　the　touch　information　of　user＇s　hands,　the　acceleration　information　and　angle　of　inclination　information　of　human　trunk,　respectively.　Then,　three　kinds　of　feature　information　were　extracted　separately　and　fused　up　by　RBF　neural　network　to　get　probability　of　elderly　fall.　If　the　probability　exceeded　the　threshold,　it　was　judged　that　the　human　body　had　a　tendency　to　fall.　Finally,　experimental　system　construction　and　verification　were　carried　out　and　the　results　showed　that　the　prediction　method　of　the　user　fall　was　reliable,　and　the　overall　prediction　accuracy　rate　was　98%.　Among　them,　the　prediction　accuracy　of　normal　samples　and　user　fall　samples　were　100%　and　96%),　respectively.　This　method　can　predict　the　user　fall　more　accurately,　and　it　provides　some　guarantees　for　preventing　elderly　fall.　©　2019　ACM.