Working memory is highly limited in its capacity. This limitation constrains how precisely we remember visual content. Integrating multiple items into chunks of information provides a way to improve memory precision. While chunking is well-established in other working memory domains, its role in visual working memory remains unclear. We tested whether people organize visual memories using geometric primitives (e.g., recognizing two horizontally-oriented gratings as forming a "line").

We reanalyzed data from two whole-report working memory experiments including ~21,000 trials. Participants memorized either 2 or 4 orientations of simultaneously presented Gabor gratings. After a short delay, they reproduced each Gabor’s orientation sequentially and rated their memory confidence. Importantly, each of the to-be-memorized Gabor orientations were randomly selected and presented at one out of six possible positions placed on a circular array around fixation per trial.

We quantified how closely the arrangement of Gabors in a trial matched predefined primitives based on geometric properties like symmetry or parallelism. Participants showed greater precision and confidence when Gabor arrangements resembled geometric primitives compared to non-structured arrangements. This benefit was replicated across experiments, independent of report order, and consistent across various similarity thresholds. Interestingly, simpler geometric primitives were associated with better memory precision.

These findings establish that geometric integration operates spontaneously in visuospatial working memory, suggesting hierarchical principles of memory organization that leverage regularity to overcome capacity limitations.