Design of experiments (DoE) has been used to optimise the accuracy of CFD predictions for diesel combustion and emissions simulations. Eight CFD simulation variables concerning grid geometry and simulation time step were used as the basis for a twenty-two point DoE analysis. The results showed that for practical CFD simulations the CFD predictions were heavily dependent on local grid distribution and the calculation timestep. From the DoE statistical models, two CFD setups were predicted to give optimal combustion and emissions results with CPU times of 11 and 44 hours, model (A) and model(B), respectively. These two CFD model setups were then used to assess the accuracy of CFD combustion and emissions predictions against a series of detailed measurements performed on a single cylinder engine fitted with a common rail fuel injection system.Comparisons with the measured data for twenty-one engine test configurations showed that model (B) delivered more consistent and more accurate predictions of in-cylinder pressure, engine power, fuel consumption, NOx and soot. Predictions of mean in-cylinder pressure were within 4% of the measured values, predicted engine power and fuel consumption were within 6%. The best predictions were within 1%. With CFD model setup (B), NOx was predicted within 10-1 5% for all, but three, of the twenty-one engine test points. Realistic NOx and soot trends were also obtained from model (B).