A　method　of　self-adaptive　impulse　dictionary　matching　pursuit　for　bearing　fault　diagnosis　was　presented　here.　Firstly,　a　new　dictionary　model　based　on　the　characteristics　of　bearing　fault　signals　was　established　and　bearing　speed,　and　size　etc.　were　introduced　into　the　dictionary.　According　to　the　effect　level　of　each　key　parameter　in　this　model　on　the　analysis　results,　the　impact　position　information　was　taken　as　the　first　key　parameter.　A　dictionary　called　the　self-adaptive　dictionary　was　built　with　the　method　of　varying　parameters　individually.　In　this　dictionary,　each　element　was　made　to　have　a　good　similarity　to　the　fault　signals　to　be　analyzed,　so　the　dictionary＇s　redundancy　was　reduced　and　its　usage　efficiency　was　lifted.　Combining　with　the　principle　of　matching　pursuit,　the　method　of　self-adaptive　impulse　dictionary　matching　pursuit　was　established.　At　last,　the　results　of　simulations　and　tests　with　this　proposed　method　showed　that　the　bearing　faults　at　different　positions　can　be　diagnosed　effectively　with　the　proposed　method;　this　method　is　better　than　the　method　of　matching　pursuit　with　genetic　algorithms.