Emotions are almost absent from scientific articles except surprise, which plays a specific role in this genre. Surprise markers such as contrairement à nos attentes (‘contrary to our expectations’) or ce résultat relativement surprenant (‘this somewhat surprising result’) are used in the framework of a scientific prediction model, implicit or explicitly formulated. A corpus linguistic study of adjectival and verbal markers allows to determine several trends: (a) contrary to other genres such as novels or newspaper articles, surprise is not polar, that is to say it is neither positive nor negative, but stylistically “neutral”, (b) surprise is far more source-oriented than experiencer-oriented, (c) surprise generally involves the reader as a witness, and contributes, with other rhetorical markers, to “interlocutive” dialogism, (d) surprise often occurs in a prefabricated discursive scenario including several steps: (i) prediction model, (ii) observed facts, (iii) expression of surprise (or absence of surprise), (iv) explanation of surprising facts. Finally, the paper questions the status of surprise as an affect in scientific writing. It seems to be more a state of consciousness associated with the observation of complex facts. In any case, it appears to be a complex state, with rich conceptual content.