The subject of this article is cognitive segmentation in translation. Based on experiments carried out in Translog, a keyboard logging program, significant differences, and also certain similarities, were observed of cognitive segmentation when data from two different subject groups and text types were compared. In the translation of a relatively easy text, novice and professional translators were found to behave fundamentally differently with respect to the size and nature of cognitive units and the speed with which they were produced. When faced with a difficult text, the behaviour in both groups was clearly affected, but some of the differences observed between novice and professional translators in the translation of the easy text were neutralized in that the professionals took over many of the features characteristic of the novices.