According to the invention, a method is provided for the compression of data which contain a plurality of information units, comprising: determining of one or more expected values for each of the information units; for each information unit, determining a coincidence of the value of the information unit with the one or more expected values; and producing output data which contain information concerning the coinciding of the values of the information units with the one or more expected values, and information concerning the values of the information units, if non-coincidences with the one or more expected values are found, in which the expected values are formed on the basis of values of information units which have already been subjected to the preceding steps for compression and form a two-dimensional context with respect to the present information unit.