The current study investigated false recognition within a paradigm (Whittlesea & Williams, 2001a) that manipulates fluency of test item processing by presenting test items at recognition in the context of a rhyming or nonrhyming nonword prime. Experiment 1 attempted to lower the salience of the role of primes at test by presenting items with primes at study. This produced a much greater bias to respond ''old'' to recognition test items preceded by rhyming primes, but only when those primes were old. Experiment 2 ruled out a cohort activation account of this large bias. Experiments 3 and 4 demonstrated that the bias to call rhyming test items ''old'' is partially mediated by attributions of fluent processing to past experience and partially mediated by the use of familiarity for the pair on an item recognition test. Taken together, these data illustrate how relations between elements in the test context can be mistaken for episodic familiarity.