In　this　paper,　a　approach　for　automatically　generating　fuzzy　rules　from　sample　patterns　is　presented.　Then　a　self-adaptive　fuzzy　neural　network　is　built　based　on　the　fuzzy　partition　which　divides　the　input　space　with　input　and　output　information.　The　salient　characteristics　of　the　self-adaptive　fuzzy　neural　networks　are:l)　structure　identification　and　parameters　estimation　are　performed　automatically　and　simultaneously;2)fuzzy　rules　can　be　recruited　or　deleted　dynamically;3)parameters　of　rules　can　be　obtained　by　evolutionary　computation.　Simulation　results　demonstrate　that　a　compact　and　high　performance　fuzzy　rule　base　can　be　constructed.　Comprehensive　comparisons　with　other　approach　show　that　the　proposed　approach　is　superior　over　other　in　terms　of　learning　efficiency　and　performance.