In this paper it is argued that monitoring for speech errors is not the same in inner speech and in overt speech. In inner speech it is meant to prevent the errors from becoming public, in overt speech to repair the damage caused by the errors. It is expected that in inner speech, but not in overt speech, more nonword errors are detected than real-word ones, and that overt repairs of errors detected in inner speech differ from overt repairs of errors detected in overt speech in that they have shorter offset-to-repair times, are spoken with raised instead of lowered intensity and pitch, and are less often accompanied by editing expressions. These hypotheses are tested against a collection of experimentally elicited spoonerisms and a collection of speech errors in spontaneous Dutch. The hypotheses are basically confirmed.