Methods for obtaining information about structure from amino acid sequence have apparently been advancing rapidly. But just what can these methods currently deliver? The following papers present the results of a large scale experiment that we have orchestrated to determine the current state of the art in protein structure prediction. We consider that the only way to objectively assess the usefulness of prediction methods is to ensure that predictions are made without any knowledge of the answers. We therefore set out to provide a framework in which a large number of such blind predictions could be made and evaluated. The procedure consisted of three parts: the collection of targets for prediction from the experimental community, the collection of predictions from the modeling community, and the assessment and discussion of the results.