This thesis addresses the problem of combustion control for automotive engines. A method is proposed to complement existing controllers, which use lookup tables based on steady-state operation. During transients, the new control method adjusts fast control variables (ignition or injection time) to compensate for deviations in the cylinder initial conditions from their optimal values due to the inherent slow processes involved. The necessary adjustments are determined from a sensitivity analysis of theoretical dynamic combustion models. The resulting open-loop controller is implementable in real time, and does not require any additional sensor or calibration. Several case studies are considered: spark ignition engines, and highly diluted diesel engines with mono-pulse and multi-pulse injection strategies. Simulations, test bench experiments and vehicle results stress the relevance of the approach.