A method to calculate a smooth electrical conductivity versus mobility plot ("mobility spectrum") from the classical magnetoconductivity tensor in heterogeneous structures with the help of a "maximum entropy principle" has been developed. In this approach the closeness of the fit and the entropy of the mobility spectrum are optimized. The spectrum is then the most probable one with the least influence of the personal bias of the investigator for any given set of experimental data and is maximally noncommittal with regard to the unmeasured data. The advantages of the maximum entropy mobility spectrum analysis as compared to the conventional mobility spectrum analysis are demonstrated using a synthetic dataset.