A 300 cc gasoline engine has been experimentally and numerically studied to compare PFI and DI operation on naturally-aspirated and turbocharged full load operating points. Experiment outlines the benefits from DI operation in terms of volumetric efficiency, fuel economy and knock propensity but also clearly indicates worse raw engine-out CO emissions. The latter is an indication of the survival of a large scale mixture heterogeneity in this downsized GDI engine even when early injection and intense induced fluid motion are combined.For such a full load operation, the application of optical diagnostics to study mixture heterogeneity cannot be considered because pressure and temperature exceed sustainable levels for transparent materials. Therefore, 3D CFD RANS computations of the intake, injection, combustion and pollutant formation processes including detailed chemistry information are performed to complement the experimental data. The results at the end of compression stroke confirm the existence of a large scale mixture stratification. Moreover, the degree of mixture stratification can be directly related to the measured CO emissions. Combustion and pollutant formation modeling with detailed chemistry information allows quantifying the influence of locally rich mixtures on flame propagation and post-flame chemistry (CO and NOx emissions).Computations for various injection timings on a unique operating point explain why the experimentally chosen injection timing is optimal. The interaction of intake-induced fluid motion with injection-induced fluid motion gives rise to a trade-off between turbulence and mixture homogeneity near TDC. Increased turbulence favors combustion speed and reduces knock propensity, which should lead to a better efficiency. But the gain in turbulence is counter-balanced by an enlarged mixture heterogeneity producing larger CO emissions and therefore impeding global efficiency.