Sparse decomposition expresses a signal as a linear combination of a small number of atoms in a redundant dictionary. The accuracy of the decomposition has an important influence on its wide application. A sparse decomposition method based on quantum evolution algorithm is proposed. The Gabor atoms are encoded by the enhanced qubit probability amplitude. Simplified gradient evolution operation and generation-by-generation reduction mutation operation are used to update the individual population. And the inner products of the residual signal of sparse decomposition and Gabor atoms are used as the fitness to filter out the best atoms for each sparse decomposition. Two experiments on sparse decomposition of simulation signals and the fault feature extraction experiments of the bearing vibration signals are carried out. It’s proved that the proposed method has a higher decomposition accuracy than the other methods.