The present experiments examined perceptual specificity effects using a rereading paradigm. Eye movements were monitored while participants read the same target word twice, in two different low-constraint sentence frames. The congruency of perceptual processing was manipulated by either presenting the target word in the same distortion typography (i.e., font) during the first and second presentations (i.e., the congruent condition), or changing the distortion typography of the word across the two presentations (i.e., the incongruent condition). Fixation times for the second presentation of the target word were shorter for the congruent condition compared to the incongruent condition, and did not differ across the incongruent condition and an additional baseline condition that employed a normal (i.e., non-distorted) typography during the first presentation and a distortion typography during the second presentation. In Experiment 1, we employed both unusual and subtle distortion typographies, and we demonstrated that the typography congruency effect (i.e., the congruent less than incongruent difference) was significant for low frequency but not for high frequency target words. In Experiment 2, the congruency effect persisted across a 1 week lag between the first and second presentations of the target words. Overall, the present demonstration of the long-term retention of superficial perceptual details (i.e., typography) supports the existence of perceptually specific memory representations. (Contains 4 tables.)