Detecting analogy is an important high-level cognitive skill that is involved in many aspects of human reasoning. While Structure Mapping Theory (Gentner, 1983) is a well-recognized high-level theory of analogy, it lacks a neural process implementation that links to perception and attention. Avoiding algorithmic computation on ungrounded symbols, we present a dynamic neural architecture built from interacting neural populations that establishes analogy between objects in two visually presented scenes. Consistent with SMT, it accounts for how humans find such analogies.