This PhD thesis, now largely of historical interest, describes a then-novel method of substructural analysis of chemical structure representations, with potential application in structure-property correlation and information retrieval within chemical information systems. Structural features are generated from Wiswesser Line Notation representations of structure, using programmes written in the COBOL language, and used as variables in multiple regression and cluster analysis applications. The methods are applied to both structurally similar and structurally diverse sets of compounds, with physical, biological and toxicological properties.