In the present paper we investigate (p, q)-directed complete bipartite graphs ?K p,q, n-directed paths ?Pn and n-directed cycles ?C n from the perspective of Granular Computing. For each model, we establish the general form of all possible indiscernibility relations, analyze the classical rough approximation functions of rough set theory and provide a close formula for the global accuracy average. Finally, we completely determine the attribute dependency function and the global dependency average for both ?C n and ?Kp,q.