A method for determining the size of each of a predetermined number of memory frames to be allocated for storing data blocks in a memory, and a processing device for implementing the method. The method comprises determining a data block size distribution for at least one set of data blocks. A data block size of the data block size distribution is eliminated iteratively until the number of data block sizes corresponds to the number of sizes of the memory frames. The data block sizes to merge are selected by means of a slack prediction algorithm. The sizes of the memory frames are determined as the data block sizes of the distribution which remain after the elimination.