GCMs incorporate deterministic chaos to reflect real-world chaotic dynamics of weather. This implies that small changes in external forcing can generate very different weather patterns, especially at local level. By using the Coupled Model Intercomparison Project phase 3 (CMIP3) multi-model dataset I show that small variations in Greenhouse Gas Emissions (GHG) and other forcing agents across the SRES scenarios generate substantial different climate scenarios in the US. By using a Ricardian model of climate change impacts on agriculture I show that the “noise” in the climate scenarios generates a “noisy” relationship between global GHG concentrations and local impacts. This implies that climate change scenarios from the CMIP3 dataset – used for the IPCC AR4 and for most of the impacts literature summarized by the IPCC AR5 – should be used with caution. This problem might be limited by providing model ensemble runs that use same initial conditions but introduce small perturbations around the central exogenous forcing scenario. With a set of exogenous forcing ensemble runs it is possible to estimate the “mean” climate pattern for each exogenous forcing scenario.