The invention discloses a multi-layer neural network based user QoE prediction method in a video service. Data is preprocessed by selecting characteristic parameters that influence user experience from the video service, and mapping the characteristic parameters into user QoE according to fault report or non fault report of the user; a QoE prediction model of the multi-layer neural network is established, and the neural network comprises an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer successively from bottom to top; the preprocessed data is input, an optimal parameter value of the model is obtained, and the established neural network model is trained; and user QoE prediction is completed; According to the invention, the data is preprocessed, important characteristic attributes are selected, different types of parameters are considered in an integrated way, the video quality predicted by the model approaches the subjective video quality experience of the user more, the user QoE can be predicted in a better way, results can be back fed timely and accurately, and a service provider and a network service provider can complete the video service as well as transmission services continuously.