Printed words that have a transposed-letter (TL) neighbor (e.g., angel has the TL neighbor angle) have been shown to be more difficult to process, in a range of paradigms, than words that do not have a TL neighbor. However, eye movement evidence suggests that this processing difficulty may occur on only a subset of trials. To investigate this issue, here we fit the ex-Gaussian distribution to naming latencies. The presence of a TL neighbor affected the [tau] parameter of the distribution of latencies, but there was no effect on the [mu] parameter. Vincentile plots confirmed that only slow responses were affected by the presence of a TL neighbor. These results contrast with the distributional effects of other variables on naming latency and are interpreted as suggesting that the TL effect involves inhibition of an incorrect output rather than broader competition between lexical entries. (Contains 4 figures, 1 table and 2 footnotes.)