Four experiments examined the Connolly & Jones (1970) model which postulates that translation between modalities in the cross-modal paradigm occurs before storage in short-term memory. In general, the results provided no support for the translation notion. Delaying until the end of the retention interval knowledge of the reproduction mode failed to produce a matching performance decrement. Subjects were able to maintain the code of original presentation through the retention interval even when they did not expect reproduction to be in this mode. In addition, the asymmetry in the cross-modal matching of visual (V) and kinaesthetic (K) information, whereby K-V performance is more accurate than V-K performance, was found to occur only under certain visual display conditions. The implications of these findings for general models of cross-modal translation were discussed.