Instrument networks for measuring surface air temperature (T) and precipitation (P) have varied considerably over the last century. Inadequate observing-station locations have produced incomplete, uneven, and biased samples of the spatial variability in climate and, in turn, terrestrial and global scale averages of T and P have been biased. New high-resolution climatologies [Legates and Willmott, 1990a; 1990b> are intensively sampled and integrated to illustrate the effects of these nontrivial sampling biases. Since station networks may not represent spatial climatic variability adequately, their ability to represent climate through time is suspect.