Using the same-different task, Perea, Dunabeitia, Pollatsek, and Carreiras (2009) showed that digits resembling letters ("leet digits"; e.g., 1 = "I", 4 = "A") primed pseudoword strings (e.g., "V35Z3D-VESZED"), but letters resembling digits ("leet letters") did not prime digit strings (e.g., "9ES7E2-935732"), and suggested that this is due to top-down feedback available for letter, but not digit, strings. Here we show that (a) single letters show as much leet priming as 3-letter words (Experiment 1); (b) leet priming is equally robust for digit strings and pseudowords when the string is 4 items long but not when 6 items long (Experiment 2); and (c) with 6-item strings, orthotactically illegal letter strings (e.g., "OIAUEQ") behave just like digit strings (Experiment 3). These results indicate that the asymmetry in leet priming is not due to top-down feedback available selectively for letter strings. We offer an alternative explanation based on the Bayesian reader account of masked priming proposed by Norris and Kinoshita (2008), and the role played by the orthotactic knowledge used to extend the functional capacity of visual working memory involved in performing the same-different task. (Contains 4 tables, 2 figures and 3 footnotes.)