In the study of recognition memory, a mirror effect is commonly observed for word frequency, with low frequency items yielding both a higher hit rate and lower false alarm rate than high frequency items. The finding that LF items consistently outperform HF items in recognition was once termed the “frequency paradox”, as LF items are less well represented in memory. However, recognition is known to be influenced both by ‘context noise’—the prior contexts in which an item has appeared—and ‘item noise’—interference from other items present within the list context. In a typical recognition list, HF items will suffer more interference than LF items. To illustrate this principle, we deliberately manipulated both the contexts in which critical items had been encountered prior to study, and the confusability of targets and distractors. Our results suggest that when noise sources are balanced, the mirror effect disappears.