This article attempts to measure the cognitive or informational load in interpreting by modelling the occurrence rate of the speech disfluency uh(m). In a corpus of 107 interpreted and 240 non-interpreted texts, informational load is operationalized in terms of four measures: delivery rate, lexical density, percentage of numerals, and average sentence length. The occurrence rate of the indicated speech disfluency was modelled using a rate model. Interpreted texts are analyzed based on the interpreter’s output and compared with the input of non-interpreted texts, and measure the effect of source text features. The results demonstrate that interpreters produce significantly more uh(m)s than non-interpreters and that this difference is mainly due to the effect of lexical density on the output side. The main source predictor of uh(m)s in the target text was shown to be the delivery rate of the source text. On a more general level of significance, the second analysis also revealed an increasing effect of the numerals in the source texts and a decreasing effect of the numerals in the target texts.