A predictive knock model is crucial for one dimensional (1-D) engine cycle simulation that has been proven to be a powerful tool in both optimization of the conceptual design and reduction of calibration efforts in development of spark-ignition (SI) engines. With application of advanced technologies such as exhaust gas recirculation (EGR) in modern SI engines, update of knock model is needed to give an acceptable prediction of knock onset. In this study, bench tests of a turbocharged gasoline SI engine with cooled EGR system operated under knocking conditions were conducted, the cylinder pressure traces were analyzed by the band-pass filtering technique, and the crank angle of knock onset was determined by the signal energy ratio (SER) and image processing method. A knock model considering multi-variable effects including pressure, temperature, EGR ratio and excess air ratio (lambda) is formulated and calibrated with the experimental data using the multi-island genetic algorithm (GA). The calculation method of the end gas temperature, the impacts of the ratio of specific heats as well as the temperature at the intake valve closure on the end gas temperature are discussed. The performance of the new model is compared with the widely-used phenomenological knock models such as Douaud-Eyzat model and Hoepke model. While the widely-used knock models fail to give acceptable predictions when increasing EGR with fuel enrichment operations, the new model predicts the knock onset with high accuracy under the fuel enrichment conditions both with and without EGR. Performance of the newly developed model is also examined in a different engine and a higher predictive accuracy of the new model than other models is obtained. (C) 2014 Elsevier Ltd. All rights reserved.