This thesis investigates the classification of chemical reactions in practice. It is motivated by the lack of discussion in the natural kind literature on the classification of reactions and other non-entity like things. I appeal to the discipline of chemoinformatics and study a new approach to the classification of reactions which is aimed specifically at meeting the needs of chemists in industry. I show that this methodology consists of three stages; the identification of a type of reaction of interest, the identification of a quantitative structural activity reaction and the importing of this data into a neural network. The output of classification is a reaction landscape which represents the similarity relations that hold between the different reactions. My aim is to outline a metaphysics that is descriptively fit for purpose with respect to my case study. I argue that such a metaphysics must be descriptively accurate, capture appropriate similarity relations and promote explanatory unification. I evaluate the entities and activities ontology proposed by Machamer, Darden and Craver, an ontology consisting on entities and dispositional properties and causal dispositionalism, against my criteria. I argue that none of these accounts are descriptively fit for purpose and that commitment to an ontological category of processes is required alongside commitment to entities and dispositional properties. I suggest that the types of reactions revealed in classification fall in the category of processes. From my analysis of reaction classification throughout the course of my thesis, I generate a list of characteristics associated with reactions and use this to provide an account of the metaphysics underlying the category of processes. My proposal focuses on the relationship between potentiality and actuality in a given chemical reaction.