A methodology to estimate CO2 emissions from an isolated power plant is presented and illustrated for the Northern Power Station at Port Augusta, South Australia. The method involves measurement of in-situ and column-averaged CO2 at a site near the power plant, forward modelling (using WRF-Chem) of the observed signals and inverse modelling to obtain an estimate of the fluxes from the power plant. By subtracting the simulated background CO2 (obtained from Monitoring Atmospheric Composition and Climate CO2 fields) from the observed and simulated signals, we are able to account for fluxes from the power plant that are mainly responsible for the variations in the CO2 concentrations. Although the enhancements of the surface concentration of CO2 are a factor of 10 larger than the enhancements in the column-averaged concentration, the forward transport model has difficulty predicting the in-situ data, which is complicated by sea breeze effects and influence from other local sources. Better simulation is obtained for the column-averaged data leading to better estimates of fluxes. The ratio of our estimated emissions to the reported values is 1.06 ± 0.54. Modelling local biospheric fluxes makes little difference either to the estimated emissions or quality of the fit to the data. Variations in the large-scale concentration field have a larger impact highlighting the importance of good boundary conditions even in the relatively homogeneous Southern Hemisphere. The estimates are insensitive to details of the calculation such as stack height or modelling of plume injection. We conclude that column-integrated measurements offer a reasonable trade-off between sensitivity and model capability for estimating point sources.