Oil well liquid level resonance signal is an important indicator fo4r oil production and transportation. However, it is often contaminated by noise, which affects the accuracy of signal analysis and processing. In order to solve this problem, a denoising method based on complete ensemble empirical mode decomposition with CEEMDAN and K-SVD is proposed in this study. Firstly, the CEEMDAN algorithm is used to decompose the original signal into a number of intrinsic mode functions (IMFs). Then, the IMFs containing noise are selected according to the correlation coefficient, and the remaining IMFs are reconstructed to obtain a denoised signal. Secondly, KSVD is employed to learn a dictionary from the noisy IMFs, and sparse coding is carried out to represent the denoised signal using the learned dictionary. Finally, the denoised signal is obtained by inverse transform of the sparse coefficients. Experiments were conducted on real oil well liquid level resonance signals, and the results show that the proposed method can effectively remove noise and retain the useful components of the signal. In addition, compared with other denoising methods, the proposed method has better denoising performance.