In this study, the authors developed an ensemble of Elman neural networks to forecast the spatial and temporal distribution of fossil-fuel emissions in 2009. The authors built and trained 29 Elman neural networks based on the monthly average grid emission data (1979–2008) from different geographical regions. A three-dimensional global chemical transport model, Goddard Earth Observing System (GEOS)-Chem, was applied to verify the effectiveness of the networks. The results showed that the networks captured the annual increasing trend and interannual variation of fossil-fuel emissions well. The difference between the simulations with the original and predicted fossil-fuel emissions ranged from ?1 ppmv to 1 ppmv globally. Meanwhile, the authors evaluated the observed and simulated North-South gradient of the atmospheric CO2 concentrations near the surface. The two simulated gradients appeared to have a similar changing pattern to the observations, with a slightly higher background CO2 concentration, ~ 1 ppmv. The results indicate that the Elman neural network is a useful tool for better understanding the spatial and temporal distribution of the atmospheric CO2 concentration and fossil-fuel emissions.