Global-land mean observations of 20th century precipitation are compared to modelled values using an optimal regression technique for nine general circulation models. The combined influence of major anthropogenic and natural forcings is detected in five cases. Comparing the accuracy of precipitation and temperature simulation of each model, we find that low temperature simulation accuracy produces low precipitation simulation accuracy, but temperature accuracy does not determine precipitation accuracy in general. Model formulation appears to be more important for accurate precipitation simulation than inclusion of a more complete set of forcings. The implications for possible constraints on land precipitation are discussed.