A model of signed language classifier predicates is presented in which these forms are held to be a mode, not of linguistic, but of visual representation. This representation is largely schematic, combining discrete parts drawn from a finite set. Some of these parts or ‘templates’ may be truly digital or undeformable in nature, but some are argued to contain ‘elastic’ parameters, allowing for the conventional use of analogue or free-form representation. The model of classifier predicates as templated visual representation thus accommodates their discrete-combinatorial structure (previously interpreted as evidence of their linguistic nature), and also accounts for the mix of fixed and nonfinite elements in them, thus solving formal problems which arise in a strictly linguistic approach. Some implications of this model include issues regarding multimodality in signed communication systems, the relationships between CPs, ‘frozen’ sign and iconic gesture, the integration of visual and abstract modes of representation, and metaphor. It is concluded that the TVR model may provide a useful new perspective on the design of representational systems in the human mind.