This reassessment of the long debate about Friedman's thesis on the pointlessness of testing assumptions in economics shows that Friedman's three famous examples, on which a large part of the credit given to this thesis is based, far from substantiating it, can be used to establish radically opposite conclusions. Furthermore, it is shown that this so-called “instrumentalist” thesis, when applied by Friedman to economics, is of a quite different nature and raises much more serious problems than the standard instrumentalist thesis devised by some methodologists of physics. To disentangle these ambiguities concerning realism and instrumentalism applied to physics or to economics, this paper refers to Van Fraassen's “constructive empiricism”, which is helpful in reformulating, in a more satisfactory way, the essentials of Friedman's considerations about empiricism and anti-realism.