Errors arising from the use of a sparse 63-station network in monitoring global and regional temperature changes are estimated by comparing results from a complete global data set from European Centre for Medium Range Forecasts analyses with those from the 63 gridpoints nearest the stations. For nine years of seasonal means, the correlations between the station network and the true values are generally quite high, but root-mean-square errors are also large outside of the tropics and are of the same order as the signal being sought. For the most part, this occurs because the variance of the 63-station network is too large so that although the overall low-frequency fluctuations are reasonably well depicted, their amplitude is too large. However, any missing data could greatly exacerbate errors arising from spatial sampling.