Network structures promote cohesive social relations among some actors and not others. Based on the theory of relational cohesion (Lawler and Yoon 1996), we hypothesize that an emotional/affective process explains how and when network structures produce such effects. The main ideas are: (1) If a network produces differential exchange frequencies among component dyads then, ceterus paribus, that network will tend to produce different degrees of internal cohesion within those dyads and will do so through the positive emotions or feelings generated by successful exchanges. (2) This effect should be more evident in equal than in unequal-power relations, and it should be weaker when network members share an overarching group identity. We conduct an experiment to test these hypotheses. The results indicate: Dyadic cohesion develops through an emotional/affective process in equal-power relations, as hypothesized, but not in unequal-power relations; and an overarching group identity reduces the degree that central actors exploit peripheral ones but does not impact dyad-level cohesion. The larger implication is that in networks containing both equal and unequal-power relations, internal pockets of cohesion are more likely to emerge in the former because of the mild, everyday positive feelings produced by successful exchanges.