The invention relates to a method in the decompression of a compressed audio signal. In the decompression, a predicting coding has been used, wherein samples taken of the audio signal have been formed into frames, and the samples of the frames have been compared with past samples to find out the prediction error. In the method, frames (fr.sub.n, fr.sub.n-1, fr.sub.n-2) of the compressed audio signal are stored, and a predicting decoding is used to decompress the audio signal compressed with the predicting coding, on the basis of said stored frames (fr.sub.n, fr.sub.n-1, fr.sub.n-2). In the method, at least one memory pointer is used to indicate the storage location of the frames (fr.sub.n, fr.sub.n-1, fr.sub.n-2). Said memory pointers (P1, P2) are used to point to the storage location of the frame (fr.sub.n-1) preceding the frame (fr.sub.n) being processed at the time, and to the storage location of the frame (fr.sub.n-2) preceding said past frame (fr.sub.n-1).