In our paper, we study data form a telecollaborative project connecting trainee teachers of French as a Foreign Language (hereafter ‘tutors’) and learners of that language (hereafter ‘learners’). In this telecollaborative project, inspired by Le français en (première) ligne project (Develotte et al. 2005), students were in contact through different online environments: a Learning Management System (Moodle), a desktop videoconference environment (Adobe Connect) and a Social Networking Site (Facebook). Following a previous study, we aim at understanding how the same pedagogical regulations (Guichon, 2012) by the tutors are accomplished in different environments, in order to understand how they adapt to the three environments. In other words, our aim is to understand which affordances of CMCL (Lamy & Hampel, 2007) are exploited to structure learning activities. More specifically, we study the techno-semio-pedagogical competence (Guichon, 2012) as it is visible through pedagogical regulations, such as giving instructions or providing positive and negative feedback (Drissi & Guichon, 2008). In our analysis, we operate a triangulation of sets of data: the pedagogical scenarios implemented in the Moodle course, the multimodal interactions within the three environments, tutors’ reflective writings about their pedagogical actions, and semi-structured interviews with the tutors. For the implementation of the Moodle course, we deployed a multimodal discourse analysis (Herring, 2004) in order to understand which modalities were used in which way. For the online interactions, either asynchronous or synchronous, we drew on multimodal conversation analysis (Develotte et al., 2011). Finally, for the reflective writings and semi-structured interviews, we employed content analysis (Bardin, 1977). A previous study focusing only on two tutors highlighted different results. Among them, we noted that those two tutors adapted their way to provide instructions making the most of the modalities available in the Moodle LMS (structure of the course, variation of format for written texts) or in Adobe Connect (with what we called ‘more conversational’ instructions). As for negative feedback, analysis showed, among others, that some way of providing feedback were not conscious. The present study deploys the same methodological setup on a wider corpus of analysis in order to find if previous findings are specific to those two tutors or if they have a more general range.