In the context of stochastic hidden variable theories, Howard has argued that the role of separability—spatially separated systems possess distinct real states—has been underestimated. Howard claims that separability is equivalent to Jarrett‘s completeness: this equivalence should imply that the Bell theorem forces us to give up either separability or locality. Howard's claim, however, is shown to be ill founded since it is based on an implausible assumption. The necessity of sharply distinguishing separability and locality is emphasized: a quantitative formulation of (...) separability, due to D'Espagnat, is reviewed and found unsatisfactory, in that it basically conflates separability and locality in a single notion. Finally, the possibility of an ‘Einsteinian’ nonseparable realism, envisaged by Shimony, is reviewed and found also to be implausible. (shrink)