The U.S. National Science Foundation’s network for monitoring UV radiation in polar regions is now in its 15th year of operation. During this period, the deployed SUV-100 spectroradiometers have repeatedly been modified, and data processing methods have been changed. These modifications have continuously improved the quality of published data, but have also introduced step-changes into the data set. For example, a change of the wavelength calibration method in 1997 has improved the wavelength accuracy to ±0.04 nm (±1σ), but also lead to a step of 2-4% in published biological dose rates. In order to best assess long-term changes in UV at network locations, it is desirable to remove these steps and to homogenize the data set. This publication discusses possible ways to accomplish these objectives, with special emphasis on absolute calibration, wavelength accuracy, and the cosine error. To date, published data are not corrected for the instruments’ cosine errors. Such corrections are not straightforward, as older data are affected by an azimuth asymmetry of the irradiance collector, which was not constant over the years. A new method to correct the errors for both clear and cloudy sky conditions was developed, and is described here. Results indicate that dose rates published prior to the year 2000 are low by 2-5%, and exhibit a variation with the Sun’s azimuth angle. By modifying the instruments’ irradiance collectors in 2000, the azimuth asymmetry was virtually eliminated, however, the modification also lead to a step-change of about 3% in published data. The ability of the new correction algorithm to remove this step is demonstrated. Uncertainties in biologically weighted dose rates caused by the cosine error can be reduced with the correction procedure to ±2%. We are planning to reprocess the entire NSF data set with the new algorithms to improve both accuracy and homogeneity.