Most developmental studies of face emotion processing show faces in isolation, in the absence of any broader context. Here we investigate two types of interactions between expression and threat contexts. First, in adults, following of another person's direction of social attention is increased when that person shows fear and the context requires vigilance for danger. We investigate whether this also occurs in children. Using a Posner-style eye-gaze cueing paradigm, we tested whether children would show greater gaze-cueing from fearful than happy expressions when the task was to be vigilant for possible dangerous animals. Testing across the 8-12-year-old age range, we found this fear priority effect was absent in the youngest children but developed to reach adult levels in the oldest children. However, even the oldest children were unable to sustain fear-prioritization when the onset of the target was delayed. Second, we addressed the development of 'threat bias' - namely faster identification of dangerous animals than safe animals - in the social context provided by expressive faces. In our non-anxious samples (i.e. with typical-population levels of anxiety), adults showed a threat bias regardless of the expression or looking direction of the just-seen cue face whereas 8-12-year-olds only showed a threat bias when the just-seen cue face displayed fear. Overall, the results argue that some, but not all, aspects of expression-context interactions are mature by 12 years of age. We measured gaze-cueing (RTinvalid-direction-trials minus RTvalid-direction-trials) and threat-bias effects from fearful (pictured) and happy faces in a context that required vigilance for danger (decide if a target animal is safe or dangerous; dangerous spider pictured). The ability to prioritize fearful-gaze in the danger-vigilance context emerged over the 8-12-year-old age range. Children also showed an adult-like threat bias for dangerous over safe animals specifically in the context of fearful faces. Overall, our results present some of the first evidence of context-expression interactions in children, and argue that studies of isolated face or threat stimuli may not apply to real-world behavior, in which contextual factors abound.