We present a novel strategy for classification of heterogeneous electron microscopy data of icosahedral virus particles. The effectiveness of the procedure, which is based on classification of single-projection reconstructions (SPRs), is first investigated using simulated data. Of several reconstruction approaches examined, best results were obtained with algebraic reconstruction techniques (ART) when providing prior information about the reconstruction in the form of a starting volume. The results presented indicate that SPR-classification is sufficiently sensitive to classify assemblies with differences of only a few percent of the total mass. The usefulness of this procedure is illustrated by application to a heterogeneous cryo-electron microscopy dataset of adenovirus mutant dl313, lacking minor coat protein IX. These data were successfully divided into two distinct classes, in agreement with gel analysis and immuno-electron microscopy results. The classes yielded a wildtype-like reconstruction and a reconstruction representing the polypeptide IX-deficient dl313 virion. As the largest difference between these volumes is found at the location previously assigned to the external portion of minor coat protein polypeptide IIIa, questions arise concerning the current adenovirus model.