Abstract: Severe harmonic noise often interferes with seismic records from vibroseis slip-sweep acquisition, greatly reducing the signal-to-noise ratio of shot gathers.Therefore, effectively suppressing harmonic noise is a precondition for high-precision and high-resolution seismic data prestack, post-stack inversion, and attribute analysis.Based on different time-frequency distributions of the effective signal and harmonic noise, this paper presents a novel method to separate harmonic noise from seismic records.The key step is the construction of redundant waveform dictionaries, which can provide a sparse representation of signal and harmonic noise.A scheme is proposed for determining the main parameters of the sparse representation dictionary.The proposed method was applied to a typical single shot record of three-dimensional seismic data in the Ma Lake area of Junggar Basin, China.The results showed that the proposed method can suppress harmonic noise effectively while maintaining good fidelity compared with conventional frequency-dependent noise attenuation (FDNAT).