Abstract: Aiming to the influence of multi-operating conditions and environmental disturbances on the state representation ability of fault features and the individual differences of fault features in sparse classification, a multi condition fault diagnosis method based on adaptive weighted sparse classifier was proposed. In this method, the score matrix for feature sparse classification performance was constructed by reconstruction residuals of each feature of the sample in the K-SVD sparse representation. Then the input feature samples were represented cooperatively and sparsely using the weight matrix obtained by iterative optimization of score matrix. The reconstruction dictionary and the sparse coefficients were updated to minimize the sparse reconstruction errors in the same mode and maximize the sparse reconstruction errors in the different modes to enhance the cooperative sparse classification performance of each feature. This method avoided the cumbersome steps of fault sensitive feature selecting and featured high-dimensional mapping in multi-operating conditions, and does not. Itdid not need a large number of fault data, either. Through the adaptive weighting of fault features and iterative optimization of sparse classifier, the established sparse dictionary can could best represent the fault state of equipment, which effectively improves improved the fault identification ability of sparse classifier in multi-operating conditions. The experimental results of rolling bearing and gearbox fault diagnosis showed that the proposed method has had higher identification accuracy and better environment robustness than the present sparse classifiers and the traditional neural network classifiers.