Database privacy is an ambiguous concept, whose meaning is usually context-dependent. We give a conceptual framework for technologies in that field in terms of three dimensions, depending on whose privacy is considered: i) respondent privacy (to avoid re-identification of patients or other individuals to whom the database records refer); ii) owner privacy (to ensure that the owner must not give away his dataset); and iii) user privacy (to preserve the privacy of queries submitted by a data user). Examples are given to clarify why these are three independent dimensions. Some of the pitfalls related to combining the privacy interests of respondents, owners and users are discussed. An assessment of database privacy technologies against the three dimensions is also included.