Combustion phasing is crucial to achieve high performance and efficiency: for gasoline engines control variables such as Spark Advance (SA), Air-to-Fuel Ratio (AFR), Variable Valve Timing (VVT), Exhaust Gas Recirculation (EGR), Tumble Flaps (TF) can influence the way heat is released. The optimal control setting can be chosen taking into account performance indicators, such as Indicated Mean Effective Pressure (IMEP), Brake Specific Fuel Consumption (BSFC), pollutant emissions, or other indexes inherent to reliability issues, such as exhaust gas temperature, or knock intensity. Given the high number of actuations, the calibration of control parameters is becoming challenging.Many different approaches can be used to reach the best calibration settings: Design Of Experiment (DOE) is a common option when many parameters influence the results, but other methodologies are in use: some of them are based on the knowledge of the controlled system behavior, by means of models that are identified during the calibration process.The paper shows how the calibration can be managed using a different concept, based on the Extremum Seeking (ES) approach. The main idea consists in changing the values of each control parameter at the same time, identifying its effect on a cost or merit function (target function), allowing to shift automatically the control setting towards the optimum solution throughout the calibration procedure. The function is evaluated cycle by cycle, based on combustion analysis. Due to the control parameters continuous variations the target function values change: the ES objective is to drive the variations towards the setting minimizing the cost function.The methodology has been applied to data referring to a GDI turbocharged engine, trying to maximize IMEP or minimize BSFC, while limiting the knock intensity and exhaust gas temperature, using SA, AFR and VVT as control variables. Experimental data referring to the considered engine have been used to feed a combustion model, allowing to test the calibration approach: results show that the ES-based calibration is able to automatically change SA, lambda and VVT values, taking into account all the constraints, and finally reaching the optimal control setting, independently of the starting setting.