Model calibration is key to model's practicality. Mathematical models, which are widely used in environmental impact assessment study to predict the quality of major components of the environment, must be calibrated and validated so as to minimize errors in the prediction. Since in most cases, model calibration needs thousands of computing, the time efficiency becomes important. As computing performance is a key challenge in model calibration, the model calibration engine in the paper adopts multi-thread technology and a multiple-machine task scheduling based on a cluster of commodity machines to improve computing efficiency greatly. An improved scheduling algorithm is proposed to ensure that the model calibration engine has a high efficiency on a cluster whose machines have different computing performance and running environment. © 2010 IEEE.