The overall aim of this thesis is to predict biological properties of molecules. The thesis first reports on the use of similarity searching for property prediction. The predictions were made by taking the value of a compound's k-nearest neighbours found from a similarity search. The initial work used structural descriptors, followed by a compound's property values (e.g. activity values across several different targets) as descriptors. The use of property value descriptors instead of classical structural descriptors showed promising results for molecular property prediction, but due to the datasets available a concrete conclusion could not be made about this technique. The use of Turbo Similarity Searching (TSS) was then investigated with the use of k-nearest neighbour predictions based on structural descriptors. . The second part of the thesis investigated the use of Free-Wilson Analysis (FWA) in conjunction with lead-optimisation and library design. It was shown that datasets can be classified into three classes: those which are successful with respect to FWA; those which are not; and those which are partially successful. For the partially successful cases it was demonstrated that it is possible to identify R-groups which do not have an independent contribution to the property being investigated. It was also found that 30% of the compounds in a full combinatorial library are sufficient to generate a successful model. Ranking the R-groups at a position on a scaffold according to their property contributions (for several different properties) can be used to generate an R-grollp profile for the R-groups, as long as a FWA is successful for the properties being considered.