Functional connectivity between the amygdala and the prefrontal cortex is critical for socioemotional processing, particularly during face processing. Though processing others' emotions is important for a myriad of complex social behaviors, more research is needed to understand how different types of emotional facial expressions differentially elicit connectivity of the amygdala with widespread neural regions. Moreover, though prior studies have reported cross-sectional associations between altered amygdala-prefrontal cortex functional connectivity and internalizing symptoms (e.g., depression, anxiety), few studies have examined whether amygdala functional connectivity is prospectively related to changes in these symptoms, with little work focusing on low-income men living in stressful contexts. The current study used psycho-physiological interaction analyses at the within-subjects level to examine how amygdala connectivity differed while participants viewed fearful, angry, and neutral faces. We used structural equation modeling at the between-subjects level, using extracted parameter estimates, to test whether amygdala connectivity during face processing predicted increases in internalizing psychopathology over time, controlling for earlier symptoms. An urban sample of 167 young men from low-income families was employed. Results indicated that negative connectivity between the amygdala and prefrontal regions was modulated by emotional face type. Neuronal activity in the cingulate and frontal cortices was connected to amygdala reactivity during fearful and neutral, but not angry, face processing. Moreover, weaker left amygdala–left middle frontal gyrus negative connectivity when viewing fearful faces and stronger right amygdala–left inferior frontal gyrus negative connectivity when viewing neutral faces at age 20 both predicted increases in internalizing behaviors from age 20 to age 22. Our findings show that amygdala-prefrontal cortex connectivity can predict the persistence of internalizing symptoms among high-risk participants over time but suggest that these patterns may differ depending on the emotional stimuli examined.