The　Guassian　distribution　model　is　often　used　to　characterize　the　statistical　behavior　of　image　or　other　multimedia　signal,　and　applied　in　fitting　probability　density　functions　of　a　signal.　But,　in　practically,　the　probability　density　function　of　data　source　may　be　inherently　non-Gaussian.　As　the　distribution　family　covers　most　of　the　common　distribution　types　and　the　frequency　curves　provided　by　the　family　are　as　wide　as　in　general　use,　this　paper　considers　Johnson　distribution　family　to　estimate　the　unknown　parameters　and　approximate　the　empirical　distribution.　The　method　uses　the　moments　to　initialize　the　parameters　of　the　distribution　family,　and　then　calculates　parameters　by　using　EM　algorithm.　The　experiment　results　show　that　the　fitted　model　could　depicts　quite　successfully　the　both　Gaussian　and　non-Gaussian　probability　density　function　of　image　intensity,　and　comparatively　the　method　has　low　computing　complexity.　©　2010　SPIE.