There is general consensus among interpreting practitioners and scholars that numbers pose particular problems in simultaneous interpreting. Adopting the view that fluency disruptions in interpreters’ renditions are signals of cognitive processing problems, the authors aim to isolate those contextual and textual factors which increase the likelihood of disfluencies when rendering numbers present in a source speech. In the reported study, we analyse data from the European Parliament Translation and Interpreting Corpus (EPTIC): we focus on target-text segments whose corresponding source segment contains a number and we find the best predictors of disfluencies by applying a generalized linear mixed model. Our approach is confirmatory and so the model accounts for factors that have been suggested in earlier studies as being associated with interpreting fluency. These factors include the nativeness of the original speaker, the type of number, the frequency of numbers in the same sentence, omission, language pair and whether the text was originally delivered impromptu or read out, and at what pace. The outcomes suggest that important predictors of disfluent renditions include omission, the frequency of numbers in a sentence and the type of number; these can be said to contribute to interpreters’ cognitive load when they process numbers.