A multivariate polynomial empirical model for evaluating the coefficient of performance (COP) of centrifugal chillers was proposed in this work. The model variables are common parameters (i.e. pressures and temperatures) on chilled water side in chiller systems. It is independent of the cooling load or partial load ratio (PLR) required by most of empirical chiller models so that it can provide a new method to evaluate the COP of a chiller without measuring the chilled water flow rate which is anyway difficult to measure directly and accurately in many on-field operating chiller plants. This makes it easy and convenient to use for on-field engineers. On-field measurement and analysis indicate that the accuracy of the proposed model is near to the compared three empirical models quoted from literatures-Gordon-Ng universal (GNU) model, Bi-quadratic regression (BQ) model, and Multivariate polynomial regression (MP) model when internal data is considered. When external data is adopted, the proposed model is obviously superior to the other three models. When internal data is used, values of the coefficient of variation of root-mean-square error (CV) of the novel model for chillers 1-3# are respectively 2.985%, 4.295%, and 2.956% with the corresponding mean relative error (MRE) values varying from 2.14% to 3.41%. While external data is adopted, the CV values of the novel model for chillers 1-3# are respectively 5.13%, 8.84%, and 4.33% with the corresponding MRE values varying from 3.19% to 6.84%. The result is acceptable for engineering applications such as building energy audit, energy conservation management, and operation management of multi parallel centrifugal chillers. (C) 2017 Elsevier B.V. All rights reserved.