Falls　in　older　people　and　the　injuries　are　a　major　problem　for　their　welfare,　confidence　and　happiness　and　represent　a　public　health　burden　on　health　care　cost.　In　this　study,　an　automatic　fall　detection　system　consisting　of　a　triaxial　accelerometer　and　a　smartphone　is　evaluated.　The　system　classifies　raw　sensor　data　by　using　an　online　algorithm.　Based　on　physical　characteristics　of　activity,　four　time-domain　features　are　abstracted,　which　are　all　independent　of　the　sensor　orientation　with　respect　to　the　body.　A　decision　tree　is　used　as　a　classifier　running　on　smartphone.　Meanwhile,　permitting　control　is　adopted　to　save　power　by　reducing　data　traffic.　The　accelerometer　and　Bluetooth　unit　are　bounded　as　a　wearable　unit　and　placed　on　the　subject＇s　waist/chest.　A　laboratory-based　trial　involving　ten　subjects　during　different　time　was　undertaken;　results　indicate　an　overall　accuracy　of　92%　and　response　time　of　less　than　6　seconds,　which　demonstrates　excellent　effectiveness　of　this　system.　©　2012　Infonomics　Society.