We often say that “A is a version of B” but do not explain what we mean by “version”. We imply that B was somehow derived from A or that they share a common ancestor. But how is B related to A? How do they differ? Do they differ in content or format? What is the significance of this difference? While this sounds like a question about the provenance of a dataset, it goes beyond that and asks questions about the identity of a digital object and the intellectual and creative work it embodies. The Research Data Alliance Data Versioning Working Group (https://www.rd-alliance.org/groups/data-versioning-ig/) collected over forty use cases of versioning practices for data and software and published a set of principles distilled from the group's analysis of them. These Principles define terminology that helps us differentiate different types of versioning and thus allow us to address the use cases more precisely. In follow-up discussions, we learned that the Principles are too abstract to apply to the operation of data repositories or to guide the citation of digital resources. Therefore, this document aims to translate the Principles into actionable recommendations for data versioning. This recommendation was developed as part of the "PID Reference Model for Versioning Research Data" (PIDsBUA) project at the research and teaching group Information Management at the Berlin School of Library and Information Science, located at Humboldt-Universität zu Berlin. The project deals with the identification and versioning of research data, which are made accessible and reusable in digital repositories, and its impact on referencing and citation in accordance with the FAIR principles, with special consideration of Open Science practices. This recommendation outlines key aspects of research data versioning for scientists and information management professionals at research-performing organisations. The recommendations are based on previous work by the Research Data Alliance Data Versioning Working Group.