Using Sternberg's paradigm, we explored short-term episodic recognition for 2-dimensional textures (summed sinusoidal gratings). The natural multidimensional representation of such stimuli makes it easy to manipulate similarity relations among study items, as well as between study and probe items. Because inter-item similarity is central to cognitive theories, controlling and manipulating similarity relations constitutes an important theoretical advance.

NEMO, our Noisy Exemplar summed-similarity MOdel, gave an excellent overall fit to recognition data from three experiments, but fell short for particular study-probe sets. Model-based analysis of individual study-probe sets revealed that participants base decisions on inter-item similarity, in addition to probe-item similarity.