The integration of heterogeneous data in varying formats and from diverse communities requires an improved understanding of the concept of a dataset, and of key related concepts, such as format, encoding, and version. Ultimately, a normative formal framework of such concepts will be needed to support the effective curation, integration, and use of shared multi-disciplinary scientific data. To prepare for the development of this framework we reviewed the definitions of dataset found in technical documentation and the scientific literature. Four basic features can be identified as common to most definitions: grouping, content, relatedness, and purpose. In this summary of our results we describe each of these features, indicating the directions a more formal analysis might take.