We consider explanations for disagreements between models and measurements of clear-sky solar transmission. Data from four cloud-free days during the FIRE 1991 field experiment are studied. Our model simulations of solar broadband fluxes exceed measurements by about 10%. About half of this discrepancy occurs in the near-infrared spectral region. About half of this discrepancy can be linked to the direct irradiance. These deviations suggest (1) model underestimates in near-infrared trace-gas absorption and (2) errors in visible diffuse fluxes. The near-infrared error in other models is generally less than the one associated with our model. Other models suggest more trace-gas absorption. However, variations among models are so large that they swamp current measurement errors such as uncertainties of the water vapor column. The visible diffuse flux error cannot be easily explained. We illustrate that a poor cosine response of solar flux instruments could have contributed. Also, underestimates of aerosol absorption could be a factor. There is clearly a need for measurements and comparisons at high spectral resolution to better quantify the sources of these errors. However, these comparisons will only be fruitful if we have confidence in the accuracy of our models and the accuracy of the data. ¿ 1998 American Geophysical Union