In humans, psychophysiological studies have shown differential influences of threat predictability on startle potentiation. Independent MRI studies show differential activations within the neural fear network, accordingly. However, joining both indices simultaneously remains a rarely accomplished goal due to the high level of artifact interference within the MRI environment. Based on our previous study, in which we established the startle probe methodology within the MRI environment, we now recorded startle EMG responses in a physiological laboratory setting and parallel to the acquisition of BOLD activity in a MRI setting within the same participants. Here, contexts were presented that differed in the degree of predictability of aversive stimuli: In the safe context (N) participants knew that no aversive stimulus was applied; in the two threatening conditions, the aversive stimulus was applied either during a visual cue (predictable, P) or at any time (unpredictable, U). Within the fear network we found stronger activations for P compared to N cues in bilateral anterior insula, left amygdala, right BNST, and ACC, whereas activity for the U context in comparison to N was enhanced only in bilateral anterior insula and right BNST. Furthermore, results show the expected potentiation of the startle response in P only in the presence of the cue, while the startle potentiation in U was independent of the presence of the cue. This pattern was identical in the laboratory and the MRI environment suggesting a high comparability of the startle probe methodology across settings.