The best‐motivated alternatives to general relativity are scalar‐tensor theories, in which the gravitational interaction is mediated by one or several scalar fields together with the usual graviton. The analysis of their various experimental constraints allows us to understand better which features of the models have actually been tested, and to suggest new observations able to discriminate between them. This talk reviews three classes of constraints on such theories, which are qualitatively different from each other: (i) solar‐system experiments; (ii) binary‐pulsar tests and future detections of gravitational waves from inspiralling binaries; and (iii) cosmological observations. While classes (i) and (ii) impose precise bounds respectively on the first and second derivatives of the matter‐scalar coupling function, (iii)a prioriallows us to reconstruct the full shapes of the functions of the scalar field defining the theory, but obviously with more uncertainties and/or more theoretical hypotheses needed. Simple arguments such as the absence of ghosts (to guarantee the stability of the field theory) nevertheless suffice to rule out a wide class of scalar‐tensor models. Some of them can be probed only if one takes simultaneously into account solar‐system and cosmological observations. © 2004 American Institute of Physics