In ratings of solar photovoltaic performance, variation in the spectrum of sunlight is commonly neglected. A single spectrum, AM1.5, is used as the sole basis not only for record laboratory efficiencies, but also for commercial module power ratings, the performance metrics for solar power plants, and warranty claims. Incorporation of solar spectrum variation would improve accuracy and reduce the financial consequences of prediction errors. Ground-level measurements of spectral irradiance collected worldwide have been pooled to provide an extensive – though by no means comprehensive – sampling of the variation. Applied to nine solar cell types, the resulting divergence in solar cell performance illustrates that a single spectrum is insufficient for comparison of cells with different spectral responses. In contrast with single-junction cells such as silicon and cadmium telluride, cells with two or more semiconductor junctions tend to have efficiencies below that obtained under AM1.5. Increases in the degree of sun tracking are shown to decrease efficiency for cells with a narrower spectral response. Of the nine cell types, silicon exhibits the least spectral sensitivity: relative site variation ranges from 1% in Lima, Peru to 14% in Edmonton, Canada, with a mean of 4%.