Our ability to link related events could be supported either byconnecting their representations in memory, or by storing themseparately but integrating their content when later drawinginferences. Here, we adapted classic memory contingencyanalyses to develop and validate an integration index designedto tap stored representations. We conducted three pre-registered experiments adopting this metric. We found positiverecall dependency for associations experienced both within thesame and across different events. Compared to a conventionalinference test, we found that recall dependency was moresensitive to a manipulation of memory integration. Leveragingrecall dependency to investigate individual differencesrevealed that better memory for contextual detail wasassociated with faster inference judgments, consistent withhigh-fidelity representations of related memories—but only forpeople who tended to store memories separately. Ourapproach, thus, provides an important tool to illuminate howrelated events are represented in memory.