Two phoneme monitoring experiments are reported that examine the amount of lexical activation produced by words containing initial, medial, or final mispronunciations. Experiment 1 showed that minimal (one distinctive feature) mismatches in the initial phoneme produced lexical activation relative to a baseline control nonword, but only when the target phoneme was situated at word offset and not word-internally. This finding suggests that considerable bottom-up support is required to override the inhibitory influence of the initial mismatching phonological information. Experiment 2 revealed no lexical activation after a medial mismatch, a finding that is consistent with bottom-up inhibition, but inconsistent with models assuming only lateral inhibition. Taken together these findings provide evidence for a selection process which includes bottom-up inhibition as a major component.