A new theoretical method has been developed for recognition and classification of membrane proteins. The method is based on computation of a polar energy surface that can reveal characteristic interaction patterns for individual helices even if crystal or NMR structure coordinates are not available. A protein with N transmembrane helices is described as a set of N vectors that are derived from a Fourier analysis of this polar energy surface computed for each helix. We then derive a polarity difference score (PDS) for any two proteins computed as the root mean square deviation between the respective vector coordinate sets. The score was found to correlate with the degree of structural similarity between the following three protein families for which tertiary structures have been determined: bacteriorhodopsin, rhodopsin, and the cytochrome c oxidase III subunit.