Within the field of SLA, studies have examined the word search practices of native and non-native speakers as interactional phenomena. Word search episodes are cases in interaction during which a speaker “displays trouble with the production of an item in an ongoing turn at talk” (Brouwer, 2003:535) and does not provide the next relevant turn but interrupts his present course of action to orient to the turn-under construction (Kurhila, 2006:20). The trouble-source can be grammatical, lexical or a longer syntactic unit. This paper focuses on lexical searches where a speaker “wishes to label a concept for which she does not have the lexical resources, or where these resources are available but cannot be recalled, or where available and retrievable resources cannot be used successfully because of contextual constraints” (Kasper & Kellerman, 1997:8). Whilst lexical searches have been examined with respect to learners’ use of different semiotic modes (Brouwer, 2004; Carroll, 2005) and strategies employed for their resolution (Kurhila, 2006), to the best of our knowledge, few studies have examined lexical word search episodes in online mediated interactions. This study builds upon previous research (Guichon & Cohen, 2014) that examined the effects of different synchronous conferencing tools on lexical word search. More specifically, it compares lexical word search in learner-teacher interactions in either a videoconferencing or audioconferencing condition. It concerns 40 undergraduate students (20 in each condition) with a B2 level in English who were asked to describe a previously unseen photograph to a native English EFL teacher. The photograph contained problematic lexical items likely to trigger word search episodes. In the videoconferencing condition, the webcam gives participants access to multimodal semiotic resources that might impact on the course of lexical searches, whereas in the audioconferencing condition, episodes rely solely on the verbal mode. The aim of this study is to explore how having access to the interlocutor’s image impacts on the lexical searches.All the interactions were recorded using dynamic screen capture software and acts in the verbal and visual modes were transcribed using ELAN (Sloetjes & Wittenburg, 2008), a multimodal data annotation tool. We chose to look in detail at one problematic lexical item, a ‘tunnel earring’, and observed the number of cases for which it triggered lexical searches. We then compared the episode duration between the two conditions, before adopting a qualitative approach to explore to what extent the information was communicated differently in each condition and what role gestures and other multimodal elements played. Finally, we examined whether one condition was more favourable to reach the target item.Initial findings suggest that the learner’s image provided by the webcam helped to launch word search trajectories as well as signal to the teacher that these were not complete and that the learner wanted to contribute further. The combination of the visual and verbal modes also allowed learners to provide more detailed and precise descriptions of the target item by contributing information in a wider range of semantic categories. The consequences of this finding in relation to language learning will be discussed.