More and more stringent emission regulations and the desire to reduce fuel consumption lead to an increasing demand for precise and close-loop combustion control of diesel engines. Cylinder pressure-based combustion control is gradually used for diesel engines in order to enhance emission robustness and reduce fuel consumption. However, it increases the cost. In this paper, a new prediction method of combustion parameters is presented for diesel engines. The experiment was carried out on a test bench to obtain the ECU (Electronic Control Unit) signals of a heavy-duty diesel engine by calibration software. The combustion parameters was measured by a combustion analyzer, such as maximum cylinder pressure (MCP), maximum combustion temperature (MCT), and combustion center of gravity (CA50). A combustion model using genetic programming (GP) is built. The input parameters are chosen from the ECU signals, such as engine speed, engine load, injection quantities, inlet air flow rate. The output parameters are MCP, MCT and CA50. The combustion model is trained and validated by measurement data. The results indicate that the combustion model can be built with the input variables of engine speed, fuel injection quantities, inlet air flow rate, inlet air temperature and exhaust air temperature. The correlation coefficient between simulation and experiment data for MCP, MCT and CA50 are over 0.90 and the average relative error is blow 4.0%.