CIF dictionaries provide a formal taxonomy of crystallographic terms and ideas. Dictionary entries are constructed in a structured machine-readable manner that facilitates validation and structuring of data. New entries may be devised for public or private dictionaries. A candidate data-name definition should fulfil the following conditions: (a) describe a specific and well defined concept: precision of definition is essential for an effective interchange mechanism; (b) have appropriate granularity: data names can define a very small piece of information (a standard uncertainty on a particular physical measurable) or a very large amount (the text of a scientific paper). An appropriate choice should be made; (c) have well defined relationships with other data items (through its assigned category membership and parent/child links); (d) constraints on the data type and permissible values should be provided where applicable; (e) the name chosen should be unique within the domain of application. This chapter describes a dictionary development strategy that meets all the above requirements. It also describes how to describe related data in multiple data blocks, how to combine and extend existing dictionaries, and how to reference and locate dictionaries against which a data file is compliant.