Most infants exhibit an attentional bias for faces and fearful facial expressions. These biases reduce toward the third year of life, but little is known about the development of the biases beyond early childhood. We used the same methodology longitudinally to assess attention disengagement patterns from nonface control pictures and faces (neutral, happy, and fearful expressions) in a large sample of children at 8, 30, and 60 months (N = 389/393/492, respectively; N = 72 for data in all three assessment; girls >45.3% in each assessment). "Face bias" was measured as a difference in disengagement probability (DP) from faces (neutral/happy) versus nonface patterns. "Fear bias" was calculated as a difference in DP for fearful versus happy/neutral faces. At group level, DPs followed a nonlinear longitudinal trajectory in all face conditions, being lowest at 8 months, highest at 30 months, and intermediate at 60 months. Face bias declined between 8 and 30 months, but did not change between 30 and 60 months. Fear bias declined linearly from 8 to 60 months. Individual differences in disengagement were generally not stable across age, but weak correlations were found in face bias between 8- and 60-month, and in DPs between 30- and 60-month (rs = 0.22-0.41). The results suggest that prioritized attention to faces--that is, a hallmark of infant cognition and a key aspect of human social behavior--follows a nonlinear trajectory in early childhood and may have only weak continuity from infancy to mid childhood.