Bearing　faults　are　a　common　cause　of　mechanical　failures.　Composite　faults　often　occur　in　bearings,　which　exhibit　mutual　interference　and　coupling　characteristics.　Traditional　intelligent　diagnosis　methods　have　limitations　in　extracting　effective　feature　information　from　composite　faults　with　a　small　number　of　samples,　leading　to　low　accuracy　and　poor　robustness.　This　paper　presents　an　intelligent　diagnosis　method　for　composite　bearing　faults　with　few　samples　to　overcome　this　limitation.　The　presented　method　is　an　end-to-end　neural　network　model　which　is　based　on　attention　mechanisms　and　feature　fusion.　Several　experiments　have　been　conducted　to　demonstrate　that　the　presented　intelligent　model　can　effectively　diagnose　composite　bearing　faults.　Furthermore,　the　effectiveness　of　individual　blocks　of　the　diagnostic　model　is　demonstrated　by　the　ablation　experiments,　and　the　proposed　model　structure　is　shown　to　be　more　effective　in　extracting　feature　information.　©　2023　IEEE.