The light source is very important when measuring solar cells. Commonly used light sources—good‐quality solar simulators—are expensive and have far from ideal characteristics. Computer modelling described in this work strongly suggests that testing of silicon solar cells under natural sunlight is simpler, cheaper, and more accurate than all but the most careful simulator measurements. Direct‐beam solar spectra were generated with the model SMARTS2 for a range of atmospheric conditions, and a broad range of silicon cells (efficiencies 6–25%) were then simulated under these spectra. These simulations showed that measurement uncertainty of less than 5% should be achievable. Climate data for locations within 45° of the equator show that the required atmospheric conditions should occur commonly in summer. Finally, it is shown that the important atmospheric conditions can be measured without expensive equipment. Copyright © 2004 John Wiley & Sons, Ltd.