Analogical reasoning is an acknowledged process behind many episodes of creativity. Typically, the creator chances upon information unrelated to the given problem – and solves the problem by analogy with this accidental source of inspiration. Current models of analogical retrieval do not explain how semantically unrelated source domains are retrieved. We present the RADAR algorithm that maps domains into a separate structure space, where domains with similar topological attributes are colocated. Each axis in structure space records the occurrence frequency of that feature in each domain. Nearest neighbour retrieval in structure space identifies structurally similar domains - from a diversity of semantic backgrounds. Structure based retrieval opens the possibility for creating an analogy model with far greater creativity potential than human reasoning.