In this paper, we investigate the process by which new experiences reactivate and potentially update old memories. Such memory reconsolidation appears dependent on the extent to which current experience deviates from what is predicted by the reactivated memory (i.e. prediction error). If prediction error is low, the reactivated memory is likely to be updated with new information. If it is high, however, a new, separate, memory is more likely to be formed. The temporal parietal junction TPJ has been shown across a broad range of content areas (attention, social cognition, decision making and episodic memory) to be sensitive to the degree to which current information violates the observer’s expectations – in other words, prediction error. In the current paper, we investigate whether the level of TPJ activation during encoding predicts if the encoded information will be used to form a new memory or update a previous memory. We find that high TPJ activation predicts new memory formation. In a secondary analysis, we examine whether reactivation strength – which we assume leads to a strong memory-based prediction – mediates the likelihood that a given individual will use new information to form a new memory rather than update a previous memory. Individuals who strongly reactivate previous memories are less likely to update them than individuals who weakly reactivate them. We interpret this outcome as indicating that strong predictions lead to high prediction error, which favors new memory formation rather than updating of a previous memory.