To understand, navigate, and interact with complex visual environments, the visual system must continuously perform intricate transformations of incoming sensory information. To enhance efficiency, it can rely on real-world regularities, such as which objects typically occur and co-occur in specific contexts. Scene grammar provides a framework to describe such regularities and the resulting complex relationships between objects in real-world scenes. To explore how scene grammar shapes information processing in the brain, we developed a graph-based representational model that encodes a scene's grammar – objects and their relationships – into low-dimensional embeddings using Graph Auto Encoders. This visual search study will investigate the alignment between neural representations, measured through EEG, and our graph-based embeddings in a flash-preview gaze contingent moving window paradigm. Using the SCEGRAM dataset, participants will search systematically manipulated scenes designed to violate either semantic relationships (object-context, i.e.: which objects are likely to appear in a scene), syntactic relationships (object-object, i.e.: the typical location of objects in scenes) or both. EEG and eye-tracking data will be collected to measure neural and behavioral responses both in anticipation of and during search. This will offer insight into how, when, and which aspects of scene grammar representations emerge to guide behavior in complex visual tasks. By comparing our structured representations to neural representations over time, Representational Similarity Analysis (RSA) may potentially reveal the timing of neural scene grammar processing in preparation of and during visual search.