An Atkinson cycle engine with geometrical compression ratio (GCR) of 12.5 has been designed by maximizing fuel economy at full load operating conditions based on the Artificial Neural Network Method [1]. However, the Atkinson cycle engine generally operates at part load conditions especially in the middle to high load range. Optimization of the fuel economy for part load is more important in reducing the total fuel consumption. The Atkinson cycle engine applies the load control strategy that combines the intake valve closure (IVC) timing and electrically throttling control (ETC), which has an impact to the fuel economy. Moreover, the exhaust valve opening (EVO) timing, spark angle (SA) and air-fuel-ratio (AFR) also affect the fuel economy. If calibrating these operating variables over the entire operating range through experiments, the difficulty and cost will become a big issue. A physical model based optimization scheme by coupling MATLAB genetic algorithm (GA) and 1-D GT-Power simulation models of the Atkinson cycle engine are proposed. The GT-Power models were improved to accurately simulate the part load conditions, by calibrating parameters of the combustion and heat transfer sub-models using experimental data taken at various speed-load points covering the entire operating range. The fuel economy was optimized based on the part-load calibrated GT-Power models using the Genetic Algorithm. After each speed-load point was optimized, the control maps for the IVC timings, SA, etc. were obtained. Then these numerically optimized control maps were input into the engine control unit (ECU) as the initial values of the engine calibration, which were further experimentally optimized. The experimental results show that the part-load GT-Power models have sufficient prediction accuracy, with maximal error of 8.5%. After optimized by GA, the fuel economy was greatly improved over the operating range, with the maximal improvement up to 7.67%. (C) 2012 Elsevier Ltd. All rights reserved.