Fragment sets generated manually from Wiswesser Line Notation have been used to correlate the chemical structures of a group of 79 penicillins with their serum binding activity, using multiple regression analysis. Statistically significant correlations were found, with results in accordance with the generally accepted nature of the binding. Algorithmic methods for the generation of such fragment sets are proposed and the use of various structural representations for structure-property correlation within chemical information systems is discussed.