It is usually claimed that in order to assess a thought experiment we should assess the nomological possibility, or realizability in principle, of its scenario. This is undoubtedly true for many TEs, such as Bohr’s reply to Einstein’s photon box. Nevertheless, in some cases, such as Maxwell’s demon, this requirement should be relaxed. Many accounts of TEs fail in this regard. In particular, experimental and some mental model accounts are too strict, since they always require realizability in principle. This paper (...) aims at analysing the notion of possibility at play in the scenarios of scientific TEs, and sheds some new light on their nature and function. (shrink)