The sources of forgetting in working memory remain the matter of intense debate. According to the SOB model (serial order in a box; Farrell & Lewandowsky, 2002), forgetting in complex span tasks does not result from temporal decay but from interference produced by the encoding of distractors that are superimposed over memory items onto a composite memory. The main tenet of the model is that the encoding strength of a distractor is a function of its novelty, with novel distractors being encoded with a large encoding weight that interferes with other memories, whereas repeated distractors would result in negligible encoding weight and no further forgetting. In the present study, we tested the 2 main predictions issuing from this model. First, recall performance should be better in complex span tasks in which distractors are repeated than in tasks in which every distractor is novel. Second, increasing the number of novel distractors should lead to more interference and poorer recall. In 5 experiments in which we controlled for attentional demand and temporal factors, none of these predictions were verified, whereas a strong effect of the pace at which distracting tasks were performed testified that they involved forgetting. We conclude that, contrary to the SOB model, the novelty of distractors plays no role per se in forgetting.