Carpenter (2011) argued that the testing effect she observed for semantically related but associatively unrelated paired associates supports the mediator effectiveness hypothesis. This hypothesis asserts that after the cue-target pair "mother-child" is learned, relative to restudying mother-child, a review test in which "mother" is used to cue the recall of "child" leads to (a) greater activation of the mediator ("father"), and (b) greater strengthening of the links in the cue-to-mediator ("mother-father") and mediator-to-target ("father-child") associative chain. This chain is then spontaneously used for recalling "child" when "mother" is given as the cue in a final test. The mediator effectiveness hypothesis is supported by the finding that relative to review restudying, "mother-child" review testing leads to better recall of the target child in the final test when cued by either mother or father. The present Experiment 1 examined an alternative account of this testing effect for mediator-to-target recall. By this account, when given as a cue, the mediator elicits the original cue, which in turn covertly cues the target via a test-strengthened cue-target association. Contrary to this account, the mediator-to-target testing effect did not depend on the preexisting mediator-cue associative strength. Experiment 2 provided a more direct test of the mediator effectiveness hypothesis by having participants recall the mediator and then the target in the final test. Contrary to predictions made by the mediator effectiveness hypothesis, (a) the cue-to-target testing effect was of the same magnitude whether the mediator was recalled or not, and (b) overall target recall was lower, not higher, when participants recalled the mediator. Thus, spontaneous mediation does not underlie the testing effect that occurs for semantically related but associatively unrelated paired associates.