A　three-stage　bearing　fault　diagnosis　method　based　on　compressed　data　and　supervised　global-local/nonlocal　discriminant　analysis　(SGLNDA)　is　proposed.　In　the　first　stage,　compressed　sensing　is　used　to　reduce　the　burden　of　storage.　The　compressed　data　can　be　obtained　from　the　original　vibration　signals　for　subsequent　fault　diagnosis.　In　the　second　stage,　a　new　manifold　learning　algorithm,　namely　SGLNDA　is　used　to　map　the　compressed　data　to　low-dimensional　space　and　retain　its　global　and　local/nonlocal　discrimination　information.　In　the　third　stage,　the　low-dimensional　features　obtained　in　the　previous　step　are　used　as　inputs　of　support　vector　machines　to　recognize　fault　diagnosis.　The　experimental　results　show　that　the　proposed　method　can　shorten　the　diagnosis　time　and　obtain　high　diagnosis　accuracy.