A semantic space model of knowledge representation in permanent memory is presented to account for metaphor processing. Concepts are represented as points at various distances along feature axes in a multidimensional space. Separate matrices within semantic space contain (1) instance-defining features, and (2) domain-relevant features. Information is mapped across semantic space by specific processing operators. High-order domains are activated first and then guide further processing, including the selective activation of instance dimensions by inference from the mapping of vehicle and topic in domain space. Results of previously reported research testing this model confirm that pragmatic principles of language must be taken into account in a model of metaphor processing to explain asymmetry resulting from reversal of vehicle and topic. Dissimilarity of topic and vehicle domains correlates with metaphor preference, and distance between topic and vehicle in domain space is predictive for vehicle selection and for decrease in topic/vehicle reversal asymmetry. (Copyright 1993, Sociological Abstracts, Inc., all rights reserved.) (J. Hitchcock in LLBA 1993, vol. 27, n. 3)