A new method for the compression of seismic sections based on a surface decomposition strategy is presented. In the decomposition process a function representing the seismic section is rewritten into a sequence of functions whose sum gives an approximation to the original data set. Data compression is obtained by keeping only the coefficient from the components that give a significant contribution to the reconstruction. By significant we shall mean that the distance between the original seismic section and the reconstruction is less than a predefined tolerance in some convenient norm. Experiments show that satisfactory interpretation quality is readily obtained with storage requirements in the range of 1 to 2 bits per sample. 10 refs., 4 figs.