Abstract Several general, empirical separation models were tested with data from a tropical site in Panama, a desert site near the Dead Sea and a temperate site in Newfoundland. The model of Bristow et al. estimated daily diffuse irradiation from total irradiation with low accuracy. This model was superior, however, to previous versions of the Liu and Jordan model for estimating monthly average daily diffuse irradiation at all but the peculiar desert site when only total irradiation was known. Fitting site-specific parameters greatly improved the accuracy of Bristow's model, but this requires diffuse radiation measurements.