Accurate forecasts of solar panel performance can improve grid penetration, enable cost evaluation prior to project implementation, and improve fault detection during operation. Spectral correction functions (SCFs) used to model the influence of the solar spectrum in such forecasting models are typically based on either proxy representations of the spectrum, using parameters such as air mass, or parameters derived directly from the spectrum, such as the average photon energy (APE). Although the latter is more accurate, the APE is argued in some studies not to be a unique characteristic of the spectrum and to suffer from increased uncertainty when analysing spectra at longer wavelengths. This study first derives APE spectral correction function coefficients for three PV technologies — multicrystalline (mSi), triple junction amorphous silicon (aSi-T), and cadmium telluride (CdTe). Based on an analysis of uncertainty in the SCF for each of the three devices, this study proposes a new spectral correction function based on the average photon energy, φ, and the depth of a water absorption band, ɛ. The additional index enables spectra to be characterised by unique combinations of φ and ɛ. Several water absorption bands are tested and the 650–670 nm band is found to yield the most accurate SCF for all three PV devices. An optimal parameterisation of the SCF for each PV device, as well as a cost-accuracy-balanced parameterisation, is presented. Improvements in the prediction accuracy of up to 60% for both the mSi and aSi-T modules, and around 20% for the CdTe module, are achieved by the proposed model with respect to a comparable two-variable proxy SCF, namely the air mass and precipitable water function. Compared with the single-variable APE SCF, f(φ), the proposed model improves the prediction accuracy by around 10% for the aSi-T and mSi modules, and by around 2% for the CdTe module. No new data are required for the proposed model compared with f(φ) as the same spectra used to calculate the APE are used to calculate ɛ. The proposed spectral correction function can easily be integrated into wider photovoltaic performance models for improved forecasting.