Solvate formation is a phenomenon that has received special attention in solid state chemistry over the past few years. This is due to its potential to both improve and impair pharmaceutical formulations. The reasons for solvate formation aren’t explicitly known. Therefore, there is currently no reliable guide in the literature on what solvents to choose in order to avoid or form a solvate when crystallizing an organic material. In this thesis we address the problem by trying to find the main reasons of solvate formation. A knowledge-based approach was used to link the molecular structure of an organic compound to its ability to form a solvate with five different solvents; these are ethanol, methanol, dichloromethane, chloroform and water. The Cambridge Structural Database (CSD) was used as a source of information for this study. A supervised machine learning method, logistic regression was found to be the optimal method for fitting these knowledge-based models. The result was one predictive model per solvent, with a success rate of 74-80 %. Each model incorporated two molecular descriptors, representing two molecular features of molecules. These are the size and branching in addition to hydrogen bonding ability. The models’ predictive ability was validated via experimental work, in which slurries of 10 pharmaceutically active ingredients were screened for solvate formation with each of the five solvents in the study. During the screening process, a new diflunisal dichloromethane solvate, a diflunisal chloroform solvate and a hymercromone methanol solvate were found. The PXRD patterns of these forms are reported. The thesis also includes SCXRD analysis of a previously known grisoefulvin dichloromethane solvate, a previously known fenofibrate polymorph and a new fenofibrate polymorph.