Data compression techniques are commonly used to achieve a low bit rate in the digital representation of signals for efficient processing, transmission, and storage. In this paper, a new technique for lossless compression of seismic data is introduced. The technique consists of two stages. The first stage is a modified linear prediction with discrete coefficients that is developed based on the assumption that the seismic data can be modeled as a sum of finite number of complex sinusoids in additive noise. In the second stage, the parameters and residual sequence of seismic data model are bi-level coded. Experimental results performed using real seismic data are presented to demonstrate the effectiveness of the new modeling approach.