To make use of the chaotic performance of ship radiated noise effectively, and apply it to feature extraction and classification of ship targets. A comprehensive feature extraction method is proposed based on phase reconstruction and correlation dimension fitting estimation, it is applied to extract ship radiated noise feature successfully. Average mutual information method and false nearest neighbors percentage method is used to select the two important parameters for phase reconstruction. Correlation dimension calculate algorithm is proposed, use least-squares curve fitting method to estimate the result precisely. The results of correlation dimension of different types of targets are simulated based on the actual measurement data, simulation results show that the feature parameter which is get by this method have nice classification performance, that it makes a good basement for the further classification and recognition of ship targets research.