This paper explores a labelling feature designed to support higher-level online dialogue. It investigates whether students use labels less often during a structured online dialogue than during an unstructured one, and looks at students' reactions to labelling and to both types of tasks. Participants are from three successive course offerings of a Master's-level course (n = 37). All students are allowed but not required to use a labelling feature which enables them to insert phrases such as "Building on your point" directly into their online messages. All students participate in two types of online activities in small groups--first an unstructured online dialogue, then a structured online dialogue. Students tended to use labels significantly less often during the structured dialogue: F(1, 36) = 5.950, p less than 0.05. Sixty-two percent of students used the feature more than once during the unstructured dialogue compared to 46% during the structured dialogue. The maximum number of labels that a student used in the unstructured dialogue was 28 versus 16 in the structured dialogue. Students generally found the structured dialogue to be more interesting and relevant, and to have clearer expectations. Student reactions to the labelling feature were mixed: The mean of satisfaction was 18.35, SD = 3.88 (six items on a 5-point Likert scale). Students did not find labelling as useful during the structured dialogue: Perhaps labelling and the activity provided redundant scaffolding. These results imply that features built into the software should be implemented flexibly with thought to the other pedagogical scaffolds in the environment, particularly to the type of activity.