Uncertainties in modeled spectral UV irradiances under cloud-free conditions are analyzed with respect to limited measurement accuracy of actual atmospheric input parameters or their nonavailability under the assumption that no uncertainty results from the used model or from the spectral values of the extraterrestrial solar irradiance and the gaseous absorption coefficients. The resulting mean uncertainty of spectral UV irradiance is calculated using a root-mean-square (rms) procedure for various scenarios, defined by differing qualities of the used sets of input values. The results are discussed with respect to the possibility of reducing the uncertainty in modeled UV irradiances by additional measurements of input parameters and, on the other hand, assessing which of such measurements may be redundant since greater measurement expense leads to no significant improvement in accuracy of modeled irradiances. The uncertainties in modeled UV irradiances are mainly produced by the uncertainties of the measured ozone amount, by the aerosol optical depth if it is not directly measured, and by the soot concentration of the aerosol in the haze layer. Additional uncertainties can arise where snow cover is present. If O3 and SO2 contents, spectral aerosol optical depth, and aerosol soot concentration near the ground are measured under actual conditions, the uncertainties in input parameters result in a mean uncertainty of about 5% for spectral integrals of UV irradiance. These results cannot be improved significantly, even when measured values of vertical profiles of all atmospheric constituents are used. Using only the observed visibility without the measurement of aerosol optical properties, the mean uncertainty for modeled UV integrals is about 10--15%.¿ 1997 American Geophysical Union