A novel computer algorithm for library design in combinatorial chemistry, GLARE (Global Library Assessment of Reagent), is used to select an optimal subset of reagents in two related libraries according to the Lipinski rule of five applied to the products. The optimized libraries show excellent compliance with the desired profiles although the original huge libraries do not. Then we show, using ten different virtual libraries, that (a) a relatively small fraction of commercially available reagents is of general use in drug/lead-like combinatorial chemistry and (b) that between 10 and 20% of the reagents are not of general use but specific to a library. This demonstrates the utility of using a product-based reagent selection method.