In most collections of segmental speech errors, exchanges are less frequent than anticipations and perseverations. However, it has been suggested that in inner speech exchanges might be more frequent than either anticipations or perseverations, because many half-way repaired errors (Yew...uhh...New York) are classified as repaired anticipations, but may equally well be half-way repaired exchanges. In this paper it is demonstrated for experimentally elicited speech errors that indeed in inner speech exchanges are more frequent than anticipations and perseverations. The predominance of exchanges can be explained by assuming a mechanism of planning and serial ordering segments during the generation of speech that is qualitatively similar to the scan-copier model proposed by Shattuck-Hufnagel (Sublexical units and suprasegmental structure in speech production planning. In P.F. MacNeilage (Ed.), "The production of speech (pp. 109-136)". New York: Springer). (Contains 5 tables and 1 figure.)