Although the MINERVA 2 model provides a detailed account of many findings related to episodic memory, its assumption that primary (working) memory is a single vector of features is unrealistic because it fails to explain the feature-binding problem—how the features of multiple items or objects can be simultaneously maintained but still differentiated. For example, the model cannot explain how the features of “red circle” and “blue square” would be maintained in primary memory in a manner that allows the color features to be correctly associated with their respective shapes. Here we propose a more plausible, biologically inspired implementation of primary memory within MINERVA 2, evaluating its performance using serial-order memory tasks and contextualizing the model’s new assumptions within the broader cognitive science and neuroscience literatures.