The primary aim of this fMRI study was to assess the proposal that negative subsequent memory effects—greater activity for later forgotten relative to later remembered study items—are localized to regions demonstrating task‐negative effects, and hence to potential components of the default mode network. Additionally, we assessed whether positive subsequent memory effects overlapped with regions demonstrating task‐positive effects. Eighteen participants were scanned while they made easy or difficult relational judgments on visually presented word pairs. Easy and hard task blocks were interleaved with fixation‐only rest periods. In the later unscanned test phase, associative recognition judgments were required on intact word pairs (studied pairs), rearranged pairs (pairs formed from words presented on different study trials) and new pairs. Subsequent memory effects were identified by contrasting the activity elicited by study pairs that went on to be correctly endorsed as intact versus incorrectly endorsed as rearranged. Task effects were identified by contrasting all study items and rest blocks. Both task‐negative and task‐positive effects were evident in widespread cortical regions and negative and positive subsequent memory effects were generally confined to task‐negative and task‐positive regions respectively. However, subsequent memory effects could be identified in only a fraction of task‐sensitive voxels and, unlike task effects, were insensitive to the difficulty manipulation. The findings for the negative subsequent memory effects are consistent with recent proposals that the default mode network is functionally heterogeneous, and suggest that these effects are not accurately characterized as reflections of the modulation of the network as a whole.