Introduction The ability to remember one’s past is a two-sided coin. It allows us to relive cherished episodes but also confronts us with past events that we would rather forget. Research over the last decade indicates that this latter side is, to some degree, under voluntary control. When people confront an unwelcome reminder of a past event, they can exclude the unwanted memory from awareness. This process, in turn, impairs retention of the suppressed memory (Anderson and Green, 2001; Hertel and Calcaterra, 2005; Anderson and Huddleston, 2011). Though recent studies have started to elucidate the neural basis of this phenomenon (Anderson et al., 2004; Depue et al., 2007; Butler and James, 2010), they all leave a fundamental question unanswered: what exactly are the neurocognitive mechanisms that underlie memory suppression? The present fMRI experiment scrutinized the existence of two possible routes to forgetting unwanted memories. Both of these putative mechanisms are hypothesized to induce forgetting by limiting momentary awareness of an unwanted memory, yet they achieve this function in fundamentally opposite ways that are mediated by different neural networks. One way to exclude a memory from awareness would be to inhibit the retrieval process directly (Bergström et al., 2009). If such direct suppression were possible, it may be mediated by a disruption of mnemonic processes supported by the hippocampus (HC), a structure known to be critical to conscious recollection (Squire, 1992; Eldridge et al., 2000; Eichenbaum et al., 2007). In support of this hypothesis, blood oxygen level-dependent (BOLD) signal in the HC is typically reduced during attempts to limit awareness of a memory compared with attempts to recall a memory (Anderson et al., 2004; Depue et al., 2007; Butler and James, 2010). Thus, these situations might recruit a direct suppression mechanism that disengages retrieval processes supported by the HC (cf. Anderson et al., 2004). At the same time, attempts to exclude a memory from awareness are associated with increased activation in right dorsolateral prefrontal cortex (DLPFC; approximating Brodmann area [BA] 46/9; Anderson et al., 2004; Depue et al., 2007; Butler and James, 2010), and a stronger recruitment of this region predicts greater subsequent forgetting of the avoided memories (Anderson et al., 2004; Depue et al., 2007). Importantly, across individuals, greater DLPFC activation correlates with decreased HC activation (Depue et al., 2007). This pattern suggests that the DLPFC may inhibit HC processing to prevent the retrieval of unwanted memories and that precluding awareness in this fashion impairs the suppressed memory traces (Anderson et al., 2004). However, it is unknown whether the activation changes in these two regions reflect such direct suppression attempts, and whether they indeed compose a functional network that supports retrieval inhibition. Here, using dynamic causal modeling, we examine the hypothesis that a negative DLPFC-HC coupling mediates such a mechanism of voluntary forgetting. The opposite way of excluding an unwanted memory from awareness would be to occupy the limited focus of awareness with another competing thought, such as another memory (Hertel and Calcaterra, 2005). Because such thought substitution requires an alternative memory to be retrieved, it would presumably engage HC processing, not disengage it. It therefore could not be based on a systemic inhibition of this structure. Instead, this mechanism requires the selection between the substitute memory and the prepotent, unwanted memory. Previous research indicates that selective retrieval can weaken competing memory traces (Anderson et al., 1994; Norman et al., 2007) and that it is supported by two prefrontal regions (Wimber et al., 2008). One of these approximates to left BA 44/9. This part of caudal PFC (cPFC) is engaged during the retrieval of weak memories in the context of stronger, interfering memories (Wimber et al., 2008; Kuhl et al., 2008). Greater activation in cPFC has also been linked to reduced proactive interference from intruding memories in working memory tasks (Nee and Jonides, 2008). Accordingly, this region may also support processes that enable substitute recall while weakening the trace of the avoided memory. The second structure, left midventrolateral PFC (mid-VLPFC; approximating posterior parts of BA 45), has been implicated in the selection of a target from among retrieved memories (Kuhl et al., 2007, 2008; Badre and Wagner, 2007). Thus, controlling awareness of unwanted memories by thought substitution may be achieved by cooperative interactions between left cPFC and mid-VLPFC that bias retrieval toward the selective recollection of distracting substitute thoughts that occupy awareness. To scrutinize the two putative mechanisms of voluntary forgetting, two groups of participants encoded reminder-memory pairs (e.g., BEACH-AFRICA). Participants then received substitute memories for a subset of these reminders (e.g., BEACH-SNORKEL) (Figure 1A). Afterward, they were scanned by fMRI while they recalled some of the associates and suppressed others (Anderson and Green, 2001). Critically, one group accomplished this in a manner likely to engage the hypothesized direct suppression mechanism. These participants attended to the reminder on the screen (e.g., BEACH) while trying to prevent retrieval of the associated memory (e.g., AFRICA). They were carefully instructed to not engage in any distracting activity (Bergström et al., 2009). If the memory entered awareness inadvertently, they were asked to block it out. By contrast, the other group performed a task likely to engage the thought-substitution mechanism, i.e., they recalled the substitute memory (e.g., SNORKEL) to help them preclude or supersede awareness of the to-be-avoided memory (e.g., AFRICA) (Hertel and Calcaterra, 2005). Afterward, we tested the mnemonic consequences of these mechanisms by probing retention of the suppressed, recalled, and baseline memories (i.e., items that were initially learned but not encountered during the suppression phase). We gauged the existence of these two opposing neurocognitive mechanisms first by examining whether they are supported by selective engagements of the hypothesized brain structures, and then by determining whether these structures compose functional networks that could mediate voluntary forgetting. Results Behavioral Results Distinct Characteristics of Direct Suppression versus Thought Substitution Debriefing confirmed that the thought substitution group predominantly controlled awareness of the unwanted memories by retrieving the substitutes (Figure 1B). The direct suppression group, by contrast, reported that they controlled awareness by focusing on the reminder as it appeared on the screen while attempting to inhibit the memory. The group differences were significant (substitute focus: t(32) = 10.59, p recall). These regressors and their interaction term (i.e., the PPI regressor) were estimated at the first level. Contrast images associated with the PPI regressor were then entered into the regression analyses at the second level. SPMs were thresholded at p < 0.05, small-volume FWE corrected for the mid-VLPFC ROI.