Surface ultraviolet (UV) radiation measurements from the Robertson-Berger (RB) meter network and existing documentation of these data were examined to determine long-term variations of UV. RB meter data from 14 sites in the United States were analyzed for trends over the period 1974–1991. A more in-depth analysis of the RB meter data, including the use of supporting geophysical data, was carried out for four of the locations. Results based on analysis of data from the 14 sites show a significant negative trend of the order of −6% per decade overall, reasonably consistent with annual trends obtained by Scotto et al.  using similar data for the period 1974–1985. However, when allowance is made for mean level shifts in the data for several of the stations around 1979, which may be due to calibration and other instrument-related problems, the resulting overall trend is found to be of the order of +2% per decade and not statistically significant. An additional trend analysis using only RB meter data since 1979 at the 14 sites is also performed and leads to overall trend results similar to those from the analysis which allows for mean level shifts in the data. The more detailed analysis of data from four of the stations for the period 1979–1991 is performed to investigate the extent to which the trend behavior in the RB meter measurements can be explained by the behavior of other geophysical quantities such as cloudiness and total ozone. In particular, radiative transfer model-based calculations of ultraviolet irradiance based on satellite data from the total ozone mapping spectrometer are compared with the RB meter measurements to help explain their behavior. Generally, inconsistencies are found between the trend behavior in RB meter measurements and radiative transfer calculations, with the RB data showing substantial downward movement relative to the calculations for three of the four sites. Significant evidence exists to indicate that problems with the network render the existing RB meter measurements unreliable for long-term trend detection. Different reasonable treatments of the data result in dramatically different trend results. Without further information, the data, by themselves, do not allow for definitive trend analysis results.