Effective feature extraction of the heat exchange tube fouling signal is the essential step for fouling thickness detection. In view of the echo energy decentralizing and model aliasing, a signal processing method based on CEEMD wavelet adaptive threshold is proposed. Firstly, the similarity between intrinsic mode function (IMF) and original signal is calculated by the angle cosine method. The signal and noise mode segmentation point is determined and evaluated combining with the energy spectrum. Besides, the wavelet adaptive threshold is used to collect detail information in noise modes. Finally, all of the remained IMFs are reconstructed to obtain a noise suppressed signal. The results show the accuracy of segmentation point is high. Improved CEEMD has better denoising performance than wavelet threshold. The numerical simulation matches the test results, proved that the proposed method is significant to extracting the feature of the thin fouling signal.