In order to model daylight availability and distribution in architectural spaces, simulation tools require reliable representations of boundary conditions - typically in terms of sky luminance distribution models However, the impact of sky model errors on simulation-based indoor illuminance predictions is not well documented. There are different tools and methods to simulate indoor illuminance conditions and related daylight indicators. In the present study, the Radiance lighting simulation program was selected. In order to generate sky scene description, Radiance contains two routines, Gendaylit and Gensky. These routines require, as input, information on both direct and diffuse components of solar radiation. To explore the implications of the sky model selection on the fidelity of simulation results, Radiance was used to compute the indoor illuminance in an existing test room on the rooftop of a university building. Therefore, the aforementioned two sky models were considered. A third option sky scanner (SC) was a sky model, generated via measured values obtained from a sky scanner. Simultaneously, the actual illuminance levels in the room were monitored under different outdoor conditions (clear, intermediate, overcast). A comparison of the measurement results, with the multiple model prediction results, facilitates an empirically based evaluation of the reliability of indoor illuminance predictions in the face of different assumptions pertaining to the prevailing boundary conditions.