As the single-signal carrier frequency estimation method is unsuitable for the time-frequency overlapped signals in an underlay cognitive radio network (CRN), in this paper, we propose a novel carrier frequency estimation method for the time-frequency overlapped multi-level amplitude-shift keying (MASK) signals in the underlay CRN. In this method, the diagonal slice spectrum of cyclic bispectrum for the time-frequency overlapped MASK signals is first estimated, and then, the carrier frequency of component MASK signals is estimated by extracting the position information of the diagonal slice spectrum line based on the norm theory and the adaptive threshold. In addition, the Cramer-Rao bound (CRB) of the carrier frequency estimation for the time-frequency overlapped MASK signals is also derived. The simulation results show that the proposed method can estimate the carrier frequency of the time-frequency overlapped MASK signals effectively, especially in low signal-to-noise ratio (SNR) regions.