This paper presents a sketch of a moderately anti-realist position in philosophy of science that is a modification of Van Fraassen’s constructive empiricism and that I call ‘adaptive empiricism’. This modification is motivated by the intuition that assessing what is or is not observable should be an important element of theory choice for an empiricist. (I use cases of underdetermination as examples.) Thus I argue that Van Fraassen’s distinction between what is observable and what is unobservable should be adapted to changing theoretical and experimental contexts. I close with some ideas as to how to develop this position more fully.