This paper presents the design and instrumentation of a 3:5-scale physical testbed of an intelligent building. The testbed is designed to enable performance evaluation of various temperature control algorithms in a controlled and repeatable setting. Key features of this testbed include fully controlled mass flow and supply air temperature, a sensor-rich environment (producing both temperature and energy measurements), control of the ambient temperature around the testbed, and a modular structure with multiple zones and varying degrees of thermal and mass flow coupling. The testbed is partitioned into 6 rooms; the interconnected structure of these rooms allows us to study the thermal coupling that occurs between adjacent zones and to explore the challenges associated with under-actuated zones. Air conditioning is provided by wall mounted thermoelectric coolers controlled wirelessly from a central computer. A unique feature of this testbed is its placement inside a larger temperature-controlled enclosure, which allows simulation of time-varying ambient weather. Precise control of ambient temperature provides a means for robust comparison and evaluation of control architectures. As a preliminary demonstration, we present experimental results comparing the performance of a decentralized proportional-integral controller and a decentralized adaptive controller under time-varying ambient temperature.