In a blind test of protein-docking algorithms, six groups used different methods to predict the structure of a protein complex. All six predicted structures were close enough to the experimental complex to be useful; nevertheless, several important details of the experimental complex were missed or only partially predicted.