The invention discloses a user abnormal behavior detection method based on neural network clustering. The method comprises: firstly, performing SVD decomposition and denoising on a behavior data set matrix of a user, and inputting the denoised matrix to an input layer of a neural network, and then calculating weight of all attributes of each user in a hidden layer of the neural network, and standardizing the weight, outputting the weight in an output layer, and finally respectively calculating similarity value of each user and each user in a normal behavior model database and a threshold value, if the similarity value is larger than the threshold value, behavior being abnormal behavior, and carrying out corresponding prompt and precautionary measures, otherwise, the behavior being normal behavior, and combining the behavior in the normal model database to upgrade the database in real time. Through cooperation of each part, the method effectively realizes high detection rate and low false alarm rate.