The exploration of structure–activity relationships (SARs) in chemical lead optimization is mostly focused on activity against single targets. Because many active compounds have the potential to act against multiple targets, achieving a sufficient degree of target selectivity often becomes a major issue during optimization. Herein we report a data analysis approach to explore compound selectivity in a systematic and quantitative manner. Sets of compounds that are active against multiple targets provide a basis for exploring structure–selectivity relationships (SSRs). Compound similarity and selectivity data are analyzed with the aid of network-like similarity graphs (NSGs), which organize molecular networks on the basis of similarity relationships and SAR index (SARI) values. For this purpose, the SARI framework has been adapted to quantify SSRs. Using sets of compounds with differential activity against four cathepsin thiol proteases, we show that SSRs can be quantitatively described and categorized. Furthermore, local SSR environments are identified, the analysis of which provides insight into compound selectivity determinants at the molecular level. These environments often contain “selectivity cliffs” formed by pairs or groups of similar compounds with significantly different selectivity. Moreover, key compounds are identified that determine characteristic features of single-target SARs and dual-target SSRs. The comparison of compounds involved in the formation of selectivity cliffs often reveals chemical modifications that render compounds target selective.