In the study of verbal memory, a critical question is the extent to which recognition is influenced by the prior contexts in which items have appeared (‘context noise’), as opposed to competition from other items present within the immediate task context (‘item noise’). In a standard recognition task, subjects study a list of words, and at test, discriminate between studied items (targets) and novel items (foils). To disentangle the contributions of context and item noise, we systematically manipulated both the contexts in which critical items had been encountered prior to study, and the composition of the recognition list, varying semantic similarity among items. Our results suggest independent contributions of each factor, with word frequency and temporal lag as important mediating variables. These findings can be interpreted within both associative learning and memory paradigms.