We describe a gapped structural threading method starting from aligning the query protein sequence to the dominant eigenvector of the structure contact-matrix. A mathematically straightforward iteration scheme provides a self-consistent optimum global sequence-structure alignment. The computational efficiency of this method makes it possible to search whole protein structure databases for structural homology without relying on sequence similarity. The sensitivity and specificity of this method are discussed, along with a case of blind test prediction. This method will provide a versatile tool for protein structure prediction and protein domain recognition complementary to existing tools that rely on sequence homology. (C) 2004 Elsevier Ltd. All rights reserved.