<p>It has been proposed that relative familiarity (i.e., the increase in familiarity relative to an item’s pre-experimental familiarity) results from an attribution process, by which the perceived discrepancy between expected and perceived fluency is attributed to the past. This mechanism potentially contributes to the higher hit rates for low- compared with high-frequency words in the word-frequency mirror effect as the fluency of a studied low-frequency word might be less expected than the fluency of a studied high-frequency word. Importantly, attributions to familiarity are more likely if the increase in fluency is not only unexpected for the given item class (e.g., low-frequency words), but also in comparison to the context (e.g., when low-frequency words are presented intermixed with high-frequency words, for which the increase in fluency is relatively small). Experiment 1 used a random and a blocked version of the two-list exclusion task combined with a short response deadline. High- and low-frequency words were studied in two modalities, one of which had to be classified as old during retrieval. We found more hits and exclusion errors (false alarms to nontargets) for low- than for high-frequency words when word frequency varied randomly, indicating that relative familiarity drove responses. However, these differences were eliminated when word frequency was blocked across study–test cycles and when a random study phase was combined with a blocked test phase (Experiment 2), indicating that relative familiarity results from an attribution process which is contingent on contextual factors during retrieval.</p>