Research on masked transposed-letter priming (i.e., jugde-JUDGE triggers a faster response than jupte-JUDGE) has become a key phenomenon to reveal how the brain encodes letter position. Recent behavioural evidence suggests that the mechanism responsible for position coding in a masked priming procedure works with familiar "object" identities (e.g., letters, digits, symbols) but not with unfamiliar object identities (e.g., pseudoletters). Here we used event-related potentials (ERPs) to explore the time course of masked transposition priming of letters vs. pseudoletters in a cue-target same-different matching task. Target stimuli were preceded by a masked prime that could be: (i) identical to the target; (ii) identical to the target except for the transposition of two internal letters/pseudoletters; or (iii) identical to the target except for the substitution of two internal letters/pseudoletters. Only cue-target 'same' trials were analyzed. The priming manipulation affected the "same" trials of the letter strings between 250 ms and 450 ms: identity and transposition conditions produced less negative amplitudes than the substitution condition. Because of the onset latency of this priming effect, we suggest that masked primes affected mainly the cognitive processes related to the categorization of the trials (match versus mismatch), rather than to the initial stages of orthographic processing.