Whether the balcony spill plume will rise unhindered as a free plume or curl inward and interact with the atrium structure is dependent upon a number of factors. Not all the factors are well investigated and wholly understood, resulting in limited guidance for fire engineers. This article systematically investigates the effects of varying balcony breadths, plume widths, and fire sizes on smoke contamination in upper balconies through a series of smoke flow experiments conducted using a one-tenth physical scale model representing a six-story atrium building. The scale model simulated a fire in an adjacent compartment connecting to a fully open atrium. Visual observations and temperature measurements of the smoke flows were carried out. From the results, it was established that the extent of smoke contamination in upper balconies increased with decreasing balcony breadths, increasing plume widths, and decreasing fire sizes. Analysis showed that the aspect ratio of plume width to balcony breadth can be used to determine whether smoke contamination of upper balconies will occur. Where contamination is likely, an empirical correlation was developed to determine the minimum height of contamination above the lowest balcony of smoke contamination.