In polymorph screening it is imperative to maximize the physicochemical diversity of solvents used in crystallization experiments. This is done in order to increase the probability of finding as many crystal forms of the compound as possible. When setting up a complete solid form screening scheme it is common practice to work with solvent subsets of varying sizes, adequately spanning the chemical space. There exists a wide array of algorithms to perform this pre-experimental selection. Prior to the implementation of these methods, the impact on the physicochemical diversity observed in the selected subsets should be fully understood. In the current study subsets of 5, 10 and 20 solvents were selected from a database of 218 organic solvents times 24 property descriptors. For this purpose four mathematically different subset selection algorithms were tested and compared: Federov D-optimal selection, Kennard-Stone selection, Principal Properties selection and a Cluster-Based selection. Federov D-optimal, Kennard-Stone and the Cluster-Based selection primarily sampled in the outer regions of solvent space, whereas the Principal Properties approach represented more of a compromising solution selecting solvents closer to the centre. In conclusion, subset selection algorithms in crystallization experimental design should be used to guarantee physicochemical diversity, though it is recommended that the computed selection is supplemented with solvents selected based on chemical knowledge of the particular compound being screened. Copyright © 2008 John Wiley & Sons, Ltd.