As the number of proteins for which a high resolution structure is known grow, it is important to classify them. A classification of protein structure would be useful for example to derive structural signatures for the protein functions. Classification requires a measure of protein structure similarity: while there are tools available to align and superpose protein structures, these tools are usually slow and not practical for large scale comparisons. In this talk, I will discuss the development of protein shape descriptors that allow fast detection of similarity between protein structures.