The heat transfer coefficients and heat flux of the billet surface in continuous casting are difficult to be measured directly by hardware measurement. Therefore, the integrated approach which combines the online temperature detection with heat transfer model of inverse problem for estimating the heat transfer coefficients and heat flux is proposed. However, the inverse problem is ill-posed and the online temperature detection is often contaminated by the error. So this paper presents the modified L-M algorithm to solve the nonlinear inverse heat transfer problem. Moreover, the SAE 1007 billet is used to illustrate the validity of this integrated approach. The heat transfer coefficients and heat flux of the billet surface are estimated and the calculated values match very well with the measured values of the billet surface temperature. The experiment results show that the modified L-M algorithm not only reduces calculated amount, but also overcomes the influence of measurement error on the inverse results. Finally, the corrected heat transfer coefficients are used to improve the accuracy of the heat transfer model, which can be applied to predict the solidified shell thickness of billet, and the predicted results are confirmed by the actual industrial data. (C) 2015 Elsevier Ltd. All rights reserved.