(A) We propose that pattern separation is supported by input from a network of brain regions and initiated outside the hippocampus in frontoparietal control, sensory/visual, and medial temporal lobe (MTL) cortical regions. (B) In particular, in one pathway (dotted arrows) cognitive control regions modulate sensory regions (after receiving early novelty signal feedback from sensory and hippocampal regions), such that unique, non-overlapping features in similar stimuli are upregulated while overlapping features are downregulated. This drives an increase in neural representation of unique features in sensory regions, creating different (or ‘separated’) representations of similar stimuli and regulating MTL cortical input. Once this signal reaches extra-hippocampal MTL cortices, such as the perirhinal cortex, domain-selective (i.e., object vs. spatial) interference resolution based on modulated features takes place before the signal reaches the hippocampus. Based on this filtered input, the hippocampus maximizes (or amplifies) the outlined feature differences in a domain-general fashion and forms unique memory traces for overlapping stimuli with more differentiated representations (i.e., the filtered input to the hippocampus drives over-representation of unique features in memory). An alternative, but not mutually exclusive, possibility (solid arrow) is that cognitive control regions directly modulate the hippocampus and its representations to enhance pattern separation in line with task demands. (C) Using apple A and apple A’ as an example of two similar inputs (where A was initially viewed and stored in memory), control regions drive separation of the two inputs in sensory regions by upregulating unique features (e.g., the leaf) and downregulating overlapping features in A’. The separation (or neural distance) of the two inputs is further increased in MTL cortical regions and then the hippocampus, creating a unique memory trace for each apple.