The experimental data analysis, feature extraction, fault detection and diagnosis are focuses in recent rocket engine study. However, in practice, because of noises and the mixing of signals due to different components, the signal-to-noise ratio (SNR) is always low, so the signal analysis and the feature extraction are quite difficult. This results in the difficulty of fault detection and diagnosis. By applying independent component analysis (ICA) to the separation of source signals from mixed signals, high quality signals have been extracted for further studies. The example reveals this method is very effective.