It is argued that among possible nonlocal hidden-variable theories a particular class (called here “crypto-nonlocal” or CN) is relatively plausible on physical grounds. CN theories have the property that (for example) the two photons emitted in an atomic cascade process are indistinguishable in their individual statistical properties from photons emitted singly, and that in the latter case the effects of nonlocality are unobservable. It is demonstrated that all CN theories are constrained by inequalities which are violated by the quantum-mechanical predictions; (...) these inequalities bear no simple relation to Bell's inequalities, and an explicit example is constructed of a CN theory which violates the latter. It is also shown that while existing experiments cannot rule out general CN theories, they do rule out (subject to a few caveats such as the usual ones concerning the well-known “loopholes”) the subclass in which the photon polarizations are linear. (shrink)