A typical agricultural experiment produces vastly more data than can be reproduced in a research paper. Thus, data provided in a paper often are not detailed enough to allow the analysis to be checked, nor can the data be reanalyzed, for example, by combining them with data from other experiments or by use of techniques developed long after the original paper was published. Limitations to reanalysis are particularly unfortunate because agricultural research increasingly deals with issues that require examining large or detailed sets of data. Journals increasingly favor publishing data in digital supplements because this improves the likelihood that papers will be cited while reducing the length of journal articles. In this paper we argue that formal methods for publishing datasets from agricultural research should be established, analogous to publishing research findings. Criteria for determining whether a dataset merits publication include originality, utility, significance, completeness, quality, and usability. Mechanisms for distributing datasets also merit attention. Ensuring long-term accessibility, respecting intellectual property rights, and managing possible corrections or modifications to datasets are among issues that require consideration.