Abstract: The strong background noise,which overwhelms effective signal,is one of common problems in seismic data. Moreover,the effective signal and random noise are difficult to be separated in the time domain. However,they may be separated in the Curvelet domain. With its multi-scale characteristic,the Curvelet transform can attenuate random noise while retain effective signal by setting a threshold to curvelet coefficient. Moreover,the SNR ratio of different scale seismic sections is introduced in the threshold function,then the effective signal and weak signal can be kept to the greatest extent by reducing the threshold in some domains with high SNR ratio. Finally,using the median filter to the residual coefficient of noise,the denoised seismic data are obtained,also the weak signal is enhanced at the same time. Compared with other methods which commonly used in the detection of weak signal processing,such as wavelet trans- form and curvelet transform using fixed threshold,the Curvelet transform using adaptive threshold describes the seismic signal more effectively,and also has obvious advantages in seismic weak signal identification and noise elimination.