Bolts are the most commonly used connectors for mechanical equipment. The stability of bolt connection plays an important role in ensuring the safe operation of mechanical equipment. It is of great significance to detect the state of bolt looseness. Aiming at the four different states of bolt loosening, a bolt looseness detection method based on variational mode decomposition (VMD) and timefrequency sensitive feature combined with least square support vector machine (LSSVM) is proposed in this paper. In order to identify the four different states of Bolt looseness, a simulation experimental platform for bolt loosening detection is built, and the vibration response data of four different states of bolt looseness are obtained by accelerometer. The time-frequency sensitive features are extracted, and the IMF component energy entropy decomposed by VMD is combined to form the sensitive multi-feature vector. The extracted multifeature vectors are combined with least square support vector machine to detect different looseness states of bolts. The recognition results are compared with the results of empirical mode decomposition ( EMD)-LSSVM and EMD multi-feature-LSSVM recognition. The recognition rate of bolt looseness detection method based on proposed VMD multi-feature in this paper is better than that of EMD-LSSVM detection method.