This　paper　presents　a　novel　method　of　applying　auto-regressive　(AR)　model　and　random　forests　to　fault　diagnosis　of　power　electronic　circuit.　AR　model　is　used　to　extract　the　features　of　the　sample　data,　realize　optimum　compressed　of　fault　sample　data,　simplify　the　data　structure　in　fault　diagnosis,　enhance　classify　speed　and　precision.　By　simulating　fault　status　of　power　electronic　circuit,　this　paper　investigates　design　details　of　random　forests　classifier　and　evaluates　its　performance.　Experimental　results　show　that　the　method　is　feasible　and　effective.　©　2009　IEEE.