Like secondary structure prediction, we believe that tertiary structure prediction should also be automatic, and hence reproducible. A fully automatic protocol called HOM FOLD, for modelling proteins by homology using a fragment-based approach, is described. This paper describes the performance of the basic algorithm. Written in the GLOBAL language, it is a simple matter to add further knowledge and expertise as new rules in the protocol. The protocol was intended to be, and is normally used as, a first step only, with more elaborate algorithms for energy refinement to be applied subsequently in the context of a higher-level protocol. However, structures in agreement within 1.80 /spl Aring/ X-ray crystallographic data can be achieved that are at least as good as that obtained by interactive modelling and refinement.