Changing trends in runoff and water balance under a warming atmosphere are a major subject of interest in recent climatic and hydrological research. Forest basins represent the most complex systems including critical hydrological processes. In this study, we investigate the relationship between annual total runoff (Q), precipitation (P), and mean temperature (T) using observed data collected from 829 (forest) site years around the world. It is shown that the strong linear relationship between annual P and Q is a function of mean T. By empirically perturbing observed annual Q and P with T, a set of ΔQ-zero lines are derived for different mean T. To evaluate the extent to which the future changes in annual P and T alter Q, the future projections of ΔP and ΔT under a warming scenario (A1B) from five coupled AOGCMs (Atmosphere-Ocean General Circulation Models) are compared with the empirical ΔQ-zero lines derived in this study. It is found that five AOGCMs show different distributions with respect to the ΔQ-zero lines, which can be attributed to the contrasting dominant sensitivities of various influencing factors to water balance partitioning among models. The knowledge gained in this empirical study is helpful to predict water resources changes under changing climate as well as to interpret hydrologic simulations in AOGCM future projections.