Physical models are commonly used to assess daylighting performance of buildings using sky simulators for purpose of research as well as practice. Recent studies have pointed out the general tendency of scale model assessments to overestimate the performance, usually expressed through work plane illuminance and daylight factor profiles, when compared to the real buildings. The cause of the discrepancy between buildings and scale models is due to several sources of experimental errors, such as modelling of building details, mocking-up of surface reflectances and glazing transmittance, as well as photometer features. To analyse the main sources of errors, a comparison of a full scale test module designed for experimentation of daylighting systems and its 1: 10 scale model, placed within identical outdoor daylighting conditions, was undertaken. Several physical parameters were studied in order to determine their impact on the daylighting performance assessment. These include the accurate mocking-up of surface reflectances, the scale model location, as well as the photometric sensor properties. The experimental study shows that large discrepancies can occur between the performance figures. They lead, on average, to a relative divergence of +60% to +105% in favor of the scale model for different points located in the side lit room. Some of these discrepancies were caused by slight differences in surface reflectances and photometer cosine responses. These discrepancies were reduced to a +30% to +35% relative divergence, by putting in the effort to carefully mock up the geometrical and photometrical features of the test module. This included a sound calibration of photometric sensors, whose cosine-response appeared at the end to be responsible for the remaining relative divergence observed between the daylighting performance figures. (c) 2005 Elsevier Ltd. All rights reserved.