The buoyancy-driven air flow and the associated energy transfer within a half-scale stairwell model have been investigated experimentally and theoretically. The experimental work comprised the larger part of the investigation. The stairwell model consisted of a lower and an upper compartment connected through the stairway. The recirculation of air was maintained by a continuous supply of heat in the lower compartment. Two different cases, referred to as closed and open non-sloping ceiling stairwells, were considered. In the former, the stairwell formed a closed system, and in the latter situation the air was allowed to enter and leave the stairwell through small openings in the lower and upper compartments, a situation which may arise in practice due to the presence of cracks. The experimental work provided detailed measurements of the velocity and temperature within the stairwell model. Hot-wire anemometers of a temperature-compensated type were used to measure the velocities, and the air temperatures were measured using platinum resistance probes. These measurements, supported by flow visualisation using smoke, provided a detailed description of the flow field. Due to the symmetry condition which existed in the stairwell, the measurements were carried out in only one-half of the stairwell. The results for both closed and open cases include the velocity and temperature profiles at the throat area (minimum area between the stairway and the lower compartment ceiling) for various distances from the side wall, mean temperatures in the upper and the lower compartments, volume and mass flows up and down the stairwell. The effect of the heat input rate on these parameters is also included. The results also include the heat losses through various surfaces bounding the system, heat and mass transfer through the stairwell joints and inlet and outlet openings, and the wall temperatures. The theoretical work was concerned with a numerical prediction of turbulent flow in two-dimensions. The k-c turbulence model, with the buoyancy terms included, was adopted. The governing equations for mass, momentum, energy and those of the turbulence model were solved using a finite-volume method. The model incorporates the SIMPLE algorithm for the derivation of pressure. The wall-function method was used for the treatment of the flow near the walls. The hybrid discretisation scheme was adopted. The predicted f low pattern was in good agreement with the pattern established by experiment. The proportion of the heat loss from the upper compartment was also in good agreement with the experiment. The maximum velocities in the throat area were underpredicted. The discrepancy between the prediction and experiment is believed to arise from shortcomings of the turbulence model, the treatment of the near-wall flow and the two-dimensionality of the numerical model.