A regional greenhouse gas flux inversion system (REFIST v1.0) is described. This paper provides a comprehensive evaluation of REFIST for three provinces in Canada that include Alberta (AB), Saskatchewan (SK) and Ontario (ON). Using year 2009 fossil fuel CO₂ CarbonTracker model results as the target, the synthetic data experiment analyses examined the impacts of the errors from the Bayesian optimisation method, inversion time span, prior flux distribution, region definition and the atmospheric transport model, as well as their interactions. The posterior fluxes were estimated by two different optimisation methods, the Markov chain Monte Carlo (MCMC) simulation and cost function minimization (CFM) methods. Increasing the number of sub-regions (unknowns) beyond "optimality" can produce unstable and unrealistic fluxes for some sub-regions, and does not yield significantly different flux estimates overall. The two optimisation methods can provide comparable, stable and realistic flux results when the transport model error is small (prior R²~0.8 with synthetic observations), but both methods present difficulty when the transport model error is large (prior R²~0.3). Stable and realistic sub-regional and monthly flux estimates for the western region of AB+SK can be obtained, but not for the eastern region of ON without excluding a poorly simulated station. This indicates a real observation-based inversion will likely work for the western region for tracers with similar temporal and spatial emission characteristics to fossil fuel CO₂ [e.g. wintertime CH₄ in Canada]. However, improvements are needed with the current inversion setup before a real inversion is performed for the eastern region.