In order to overcome the complex influences of uncertain factors of information high dimension, redundancy and noise etc. in the multi-source data fusion on the network security situation evaluation, presents a network security situation assessment method based on rough set and neural network. This method uses rough set theory capabilities in machine learning, redundant information processing and feature extraction and combines with the neural network ability of dealing with noise and arbitrary approximation to construct the situation assessing model composed of index layer, discrete layer, rule layer and decision layer, and compares and studies it with BP neural network method. The simulation experiment shows that the method has less deviation and can analyze network security situation more objectively and accurately.