Attention/likelihood theory is a model of recognition memory designed to explain the mirror effect (Glanzer & Adams, 1985, 1990). The theory and the effect were studied using speed versus accuracy instructions and short versus long exposure of stimuli. Speed versus accuracy instructions during test and short versus long exposure of stimuli during study were used to vary the number of features sampled from stimuli. When the number of features sampled was reduced either by speed instructions or by shorter exposure, recognition performance was impaired. The theory predicts that in such cases, all distances between underlying distributions will contract. That means, moreover, that when recognition accuracy is decreased for old stimuli, it is also decreased for new stimuli. These predictions were supported by the data from three experiments.