Processing models of simultaneous interpreting (SI) have predicted that the linguistic complexity of the source-language text will adversely affect performance quality due to the increased storage and processing demands imposed by linguistically complicated constructions on a limited-capacity working memory (WM). The present paper reports an experimental expert-novice comparison study which examines this hypothesis by analysing the effect of four complexity metrics, validated in psycholinguistic research on interpreting performance: developmental level, amount of embedding, type of embedding and clause type. The results show a consistent effect of complexity metrics on novice performance, while for experts the effect was modulated by the overall level of textual redundancy. The findings suggest that syntactic complexity, as indexed by the measures used in this study, does not predict experts’ accuracy in a significant way. In line with Ericsson and Kintch’s theory of skilled WM, this indicates the somewhat diminished role of surface structure cues in skilled SI processing and suggests a strong component of strategic and macro processing in expert SI performance.