We have recently developed a novel strategy for the rational design of compounds. This 'in silico screening' approach is based on the design and screening of virtual combinatorial libraries. Screening is performed using defined rules derived from a comprehensive description of active and inactive molecules in a relevant learning set. This strategy allows the development of potential ligands without the necessity of any knowledge of the 3D-structure of the target receptor. Key to the success of such methods is the quality of the information being processed, in particular, the diversity of the data in the context of the molecular population in the libraries concerned. Here, we review the problem of data diversity, its definition and its analysis using a new software tool, named Diverser.