A knowledge of the three-dimensional structure of proteins is an essential prerequisite for the design of new molecules. When the tertiary structure is not available from high-resolution X-ray or n.m.r. analysis, the success of prediction is improved by using a relational database of known protein structures. This can be searched to provide information on secondary structure motifs and domains which are recognized by characteristic sequence patterns and which are assembled as `spare parts' by using computer graphics. Similar techniques can be used to give approximate structures for amino-acid replacements, deletions and insertions introduced by mutagenesis. The resulting structures are optimized by using interactive graphics, energy minimization and molecular dynamics.