In eyewitness research the frequent use of video playback presented on a computer screen (i.e., 2D videos) in laboratory-based research is problematic due to the low realism of this method when presenting, for example, threatening (and non-threatening) first-person (and third-person) scenarios. However, in contrast to 2D videos, 360-degree videos presented in virtual reality (VR) presents the opportunity of achieving more realistic and immersive scenarios that might be better suited to mimic real-life incidents, as for example, in the case of a threatening first-person robbery. In the present study, we asked 37 participants to watch eight pre-recorded threatening or non-threatening 2D and VR videos, viewed from either a first- or third-person perspective. After each video, participants assessed the observed target’s appearance and were then presented with either a target present (TP) or target absent (TA) six-person photograph line-up. We expected that VR would result in higher degrees of accuracy in both TP and TA line-ups compared with 2D and that the differences between manipulations would be more pronounced within VR compared with 2D. We found that TP (but not TA) accuracy was higher in 2D compared with VR videos (91 vs. 66%), that there was no main effect of perspective, and that threatening scenes increased TP (but not TA) accuracy compared to non-threatening scenes (86 vs. 70%). Furthermore, in VR (but not in 2D), threatening scenes increased TP (but not TA) accuracy compared with non-threatening scenes (85 vs. 40%). The results go against the expected increased accuracy in VR (vs. 2D) videos but support the notion that threatening (vs. non-threatening) scenes can increase identification accuracy in VR but not necessarily in 2D.