Threat bias, or exaggerated selective attention to threat, is considered a key neurocognitive factor in the etiology and maintenance of pediatric anxiety disorders. However, upon closer examination of the literature, there is greater heterogeneity in threat-related attentional biases than typically acknowledged. This is likely impacting progress that can be made in terms of interventions focused on modifying this bias and reducing anxiety, namely attention bias modification training. We suggest that the field may need to “take a step back” from developing interventions and focus research efforts on improving the methodology of studying attention bias itself, particularly in a developmental context. We summarize a neurocognitive model that addresses the issue of heterogeneity by broadly incorporating biases toward and away from threat, linking this variation to key neurodevelopmental factors, and providing a basis for future research aimed at improving the utility of threat bias measures and interventions in clinical practice.