The invention provides an estimation method of large-scale IP traffic in a time-frequency synchronization application, and relates to the technical field of the network management under time-frequencysynchronization network environment. The method proposes the method for estimating the large IP traffic matrix by adopting a RBF neural network based on the estimation method of the large-scale IP traffic matrix based on the RBF neural network. An estimation model of the large-scale IP traffic matrix is established by training the RBF neural network; and the good estimation on the large-scale IPtraffic matrix is realized by combining the model with the iteration proportional fitting process, and the large-scale IP traffic matrix can be dynamically tracked.