This paper aims to design decentralized iterative learning control (ILC) for building temperature control system (BTCS). The BTCS is described by a large-scale interconnected dynamic equation and modeled as a linear multiagent system subjected to undirected communication topology using graph theory. Typically, there are two types of control strategies for BTCS, namely, distributed and decentralized schemes. The main idea of designing for the decentralized scheme is to separate the whole system into a number of subsystems. In this research, we formulate and apply decentralized ILC to a four-connected-room model. In particular, we design D-type ILC for each room in the building separately, and each controller does not communicate with other controllers. The main task of the local controller is to achieve the tracking objective, i.e., the temperature of each room tracks its own desired reference temperature. Finally, numerical results illustrate the effectiveness of decentralized ILC and are compared with that of distributed consensus controller (DCC). The results show that output responses of both controllers can track trapesoidal reference and consume the same amount of total power at steady state. Decentralized ILC gives response without overshoot, and its convergence is faster than that of DCC. Convergence analysis reveals that tracking speed depends on the choice of learning gain.