Interpolated data sets are often considered to be a reliable source of information on a variety of meteorological variables, such as temperature and precipitation. Users expect the interpolated data to be rather similar to those directly observed at stations, which is not always true: well documented is the influence of interpolation on, e.g., extremes. Here another kind of discrepancy between gridpoints and station observations is presented: the time evolution of relationships between temperature and atmospheric circulation. One of the most widely utilized gridded temperature data sets, CRU TS (Climatic Research Unit gridded Time Series), is compared with 634 station time series from GHCN (Global Historical Climatology Network) in the Northern Extratropics. We analyze running correlations (calculated for 15‐year windows) of monthly values between modes of atmospheric circulation variability (identified in the ERA‐40 reanalysis) and temperature anomalies in winter from 1957 to 2002. The smallest differences in the running correlations are found in Europe and North America due to a dense station network. On the other hand, the sites with considerable differences are located mainly in mountainous regions or in isolated locations. In order to uncover causes of these differences, we analyze two sites in more detail. Mike (the North Sea) is an isolated site where the gridpoint temperature is affected by rather distant Scandinavian stations. At Songpan (central China; 2,852 m a.s.l), the terrain configuration in mountainous region influences the gridpoint value, in which the effect of stations with much lower altitude and different climate conditions is dominant.