Six experiments addressed the encoding of location information during automatization, to test a critical prediction of the instance theory of automaticity (G. D. Logan, 1988). Subjects searched 1- or 2-word displays for members of a target category. Specific targets appeared in the same locations consistently throughout training, and then location changed at transfer. Sensitivity to changes in location were assessed with implicit and explicit memory tests. In both tests, sensitivity depended on the number of locations the words could occupy (2 vs. 16). Sensitivity varied with the number of words presented (1 vs. 2) in the implicit test, but not in the explicit test. The results suggest that subjects encoded the locations of the words during automatization, which confirms the predictions of the instance theory.