A　new　method　of　fault　feature　extraction　for　rolling　bearings　based　on　improved　switching　Kalman　filter　is　proposed.　Compared　with　the　traditional　Kalman　filter　algorithm,　this　method　only　needs　the　current　measurement　value　and　the　optimal　estimation　of　the　previous　moment　in　each　iteration,　so　it　has　high　computational　efficiency　and　strong　real-time　performance.　Firstly,　the　vibration　signals　of　fault　bearings　are　divided　into　two　parts:　Fault　impulse　vibration　and　normal　vibration.　Secondly,　the　Kalman　filter　model　based　on　the　dynamic　impulse　response　of　the　bearing　mass-spring-damper　system　and　the　linear　Kalman　filter　model　are　established　respectively　for　the　fault　impulse　vibration　and　the　normal　vibration.　Then,　the　state　estimation　of　vibration　signals　is　carried　out　by　using　the　switching　Kalman　filter　algorithm　based　on　Bayesian　estimation.　Finally,　the　bearing　fault　feature　extraction　is　realized　by　filtering　noise　and　identifying　fault　impulse　components　through　time　domain　iteration　filtering.　The　simulation　and　experimental　results　show　the　feasibility　and　effectiveness　of　the　proposed　method.　©　2019　Journal　of　Mechanical　Engineering.