Recent studies on attachment in middle childhood suggest that dismissing children tend to underreport their psychological distress relative to physiological indices of distress. However, this has yet to be examined in the context of behavioral indicators of distress. In this longitudinal study, a community sample of children (N = 34, M age = 9.59 years) completed the Child Attachment Interview at Time 1. Three years later, they returned and completed the Trier Social Stress Task for Children (TSST-C), providing ratings of their state anxiety before and after the stressor. Four raters coded children’s behavioral signs of anxiety (e.g., non-signaling gestures, eye movement, posture, facial expression) on a 7-point scale during the storytelling and arithmetic tasks in the TSST-C. Hierarchical linear regressions revealed that greater attachment dismissal was associated with greater behavioral anxiety in both the story and math tasks, but was not associated with greater increases in self-reported anxiety. Dismissal prospectively predicted increased divergence of behavioral and self-reported anxiety, such that higher dismissal was associated with higher divergence scores (i.e., underreporting of anxiety relative to behavioral indicators). We discuss these results in terms of their contribution to understanding attachment and emotion in an understudied developmental phase.