Adaptive beamformer analyses of magnetoencephalograms (MEG) have shown promise as a method for functional imaging of cortical processes. Although recent evidence is encouraging, it is unclear whether these methods can both localize and reconstruct the time course of activity in subcortical structures such as the amygdala. Fourteen healthy participants (7 women) performed a perceptual matching task of negative emotional faces (angry and fearful) and geometric shapes that was designed for functional magnetic resonance imaging (fMRI) studies to maximize amygdala activation. Neuromagnetic data were collected with a 275-channel whole-head magnetometer, and event-related adaptive beamformer analyses were conducted to estimate broadband evoked responses to faces and shapes across the whole brain in 7 mm steps. Group analyses revealed greater left amygdala activity to faces over shapes, both when face-matching and shape-matching trials were presented in separate blocks and when they were randomly intermixed. This finding was replicated in a second experiment with 7 new participants (3 women). Virtual sensor time series showed clear evoked responses in the left amygdala and left fusiform gyrus in both runs and experiments. We conclude that amygdala activity can be resolved from MEGs with adaptive beamformers with temporal resolution superior to other neuroimaging modalities. This demonstration should encourage the use of MEG for elucidating functional networks mediating fear-related neural phenomena that likely unfold rapidly in time across cortical and subcortical structures.