Natural ventilation is growing in popularity as a low-energy cooling strategy. Many buoyancy-driven naturally ventilated buildings rely on atria to provide a flow path through the building. The proper design of such atria relies on validated modeling techniques. Although a plethora of studies on such techniques exist, very few rely on full-scale experiments within an atrium for validation. Even fewer studies use a naturally ventilated atrium. This study uses a full-scale naturally ventilated atrium to validate three CFD turbulence models, RNG k-epsilon, k-epsilon, and LES, all of which are found to predict experimental temperatures with an RMSE below 1.2 C. An airflow visualization technique utilizing neutrally buoyant helium bubbles is shown to be an effective method of visualizing the airflow and providing local air velocities through particle image velocimetry techniques. The LES model best predicts local velocities near the heat source. Due to an opposing wind at the roof exit, an unexpected bulk downward flow is observed through the atrium, suppressing the plume from the heat source. This downward flow highlights the necessity of using accurate modeling techniques and boundary conditions in atria design. (C) 2013 Elsevier B.V. All rights reserved.