Abstract: In order to achieve data compression and denoising of realtime collected underwater acoustic signal, this paper presents a dynamic threshold orthogonal matching pursuit (DTOMP) method. This method uses sparse representation for underwater signal pre-processing by applying threshold to greedy algorithm, meanwhile divides noise into two parts according to its characteristics to control preset parameters. Experimental researches on noise reduction of sinusoidal signal plus Gaussian noise, whale blows and ship radiated noise signal indicate that this method could improve the SNR and meanwhile compress original signal. Moreover, this method has better performance over wavelet denoising in wider dynamic range of SNR.