Monthly means of UV erythemal dose at ground level from the Robertson-Berger (RB) sunburn meter (1976--1992) and the UV-Biometer model 501 MED meter (1993--1994) located at Belsk (21 ¿E, 52 ¿N), Poland, are examined. The monthly means are calculated from all-sky daily means of UV erythemal dose. Ancillary measurements of column ozone (by Dobson spectrophotometer), sunshine duration (by Campbell-Stokes heliograph), and total (sun and sky) radiation (by a pyranometer) are considered to explain variations in the UV data. A multiple regression model is proposed to study trends in the UV data. The model accounts for the UV erythemal dose changes induced by total ozone, sunshine duration (surrogate for cloud cover variations), or total solar radiation (surrogate for combined cloud cover and atmospheric turbidity impact on the UV radiation), trends due to instrument drift, step changes in the data, and serial correlations. A strong relationship between monthly all-sky UV erythemal dose changes and total ozone (and total solar radiation) is found. Calculations show that an erythemal radiative amplification factor (RAF) due to ozone under all skies is close to its clear-sky value (about 1). However, the model gives evidence that the RAF due to ozone is smaller for cloudier (and/or more turbid) atmospheres than long-term reference. Total solar radiation change of 1% is associated with a change of 0.7% in the UV erythemal dose. Modeled trends in the Belsk's UV data, inferred from the model using ozone and total solar radiation as the UV forcing factors, are 2.3%¿0.4% (1&sgr;) per decade in the period 1976--1994. The large increase in the UV erythemal dose, of the order of 4% per decade due to ozone depletion (-3.2% per decade), is partially compensated by a decreasing tendency (-2.8% per decade) in total solar radiation. The model estimates the trend in the UV data of the order of 0.1% per decade (not statistically significant) due to superposition of the instrument drift and long-term effects related to other UV influencing factors (not parameterized by the model). ¿ American Geophysical Union 1996