One of the sources contributing to the overall uncertainty of spectral UV radiation measurements is the cosine error of the spectroradiometer. It leads to measurement errors that depend on atmospheric conditions and on solar zenith angle, and thus time of the day and season. Though the foreoptics of modern instruments are designed such as to minimize cosine errors, there remain deviations from the ideal cosine response. We have worked out a method to further reduce that remaining cosine error in global spectral irradiance. This method was applied to spectra of global UV radiation taken with a Brewer spectroradiometer. The only additional input data needed to apply the method of cosine correction to spectral irradiance data are concurrent broad-band UV-B radiation measurements of diffuse and global radiation recorded with filter UV instruments, which are used to estimate the optical thickness referred to global UV radiation for the time when the spectral scan is taken. The method takes account of the variable conditions of cloudiness and turbidity. In the case of measurements taken with Brewer instrument No. 30, the cosine corrected global UV-B radiation was higher than the measured irradiance by 9-20%, and even its daily totals turned out to be higher than the uncorrected radiation by 13-18%. An estimate of the uncertainty of +/-4 to +/-8% was derived from a theoretical approach as well as from model calculations using a radiative transfer model.