The analysis of film discourse from a multimodal and cognitive perspective has shown in recent years that such an approach to the study of cinema is a very fruitful one. Among the various cinematic techniques that may be analysed as pieces of multimodal discourse, the flashback seems to be particularly appealing because, while being very rich and versatile, it is also a fixed device and common enough in film as to be studied in a systematic way. Given those characteristics - formal variety alongside stability - a relevant question would be: how do spectators make sense of film retrospections? To address this question, this paper suggests an examination of the multimodal cues offered by flashbacks in three different films - Ordinary People (1980), Big Fish (2003) and The Help (2011) - and analyses the cognitive processes that those cues activate and which make the comprehension of the flashback possible. What lies at the basis of the flashback scenes proposed is a joint-attention triangle formed by the viewer and the camera, who look together, first at the character in the present and then at the events taking place in the past. Ultimately, such scenes can only be understood in terms of blended joint attention, and they also reveal the importance of other cognitive processes at work, namely time compression, viewpoint integration, and identity and analogy connections.