One of the most difficult problems in modern biochemistry is that of accurately predicting a protein's three dimensional structure from its sequence (the protein folding problem). This structure is essential for a proper understanding of how a protein functions. As experimental derivation of a protein's structure is far more time consuming than deriving a protein's sequence, prediction of structure from sequence is an important goal for many protein biochemists; several methods have been suggested for this. Given a protein of known structure of similar sequence to the protein you wish to model homology modelling is the method most likely to produce a fairly good model. In this work a tool was produced for examining the various stages of homology modelling and analysing how well various method for carrying out these stages perform. The tool produced consists of an object-oriented database of protein structures and testbed software written in a mixture of PROLOG and DAPLEX. Tests were carried out using this software to examine the predictivity of various guidelines suggested in the literature for the loop selection stage of cut and paste homology modelling. The results of these tests produced surprising new information on the relative importance of different factors which may be used to choose between candidate fragments for the variable regions of a protein being modelled. The results of the application of these automated modelling methods were then compared with a short series of modelling tests using human modellers in an attempt to measure how the usual modelling procedures using 'hand and eye' compare with automated measures. Finally the results of the tests carried out were used to guide the production of a model of a previously unmodelled serine proteinase.