Model-based, three-dimensional (3D) image reconstruction procedures require a starting model to initiate data analysis. We have designed an ab initio method, which we call the random model (RM) method, that automatically generates models to initiate structural analysis of icosahedral viruses imaged by cryo-electron microscopy. The robustness of the RM procedure was demonstrated on experimental sets of images for five representative viruses. The RM method also provides a straightforward way to generate unbiased starting models to derive independent 3D reconstructions and obtain a more reliable assessment of resolution. The fundamental scheme embodied in the RM method Should be relatively easy to integrate into other icosahedral software packages. (c) 2006 Elsevier Inc. All rights reserved.