The iron ore sintering process is a complex object with the characteristics of uncertainty, multivariable coupling, time-varying and time-delay. The burning-through-point (BTP）is affected by many factors and difficult to be controlled to the required precision by conventional control methods. A BTP prediction method is proposed by using the excellent time series prediction performance of support vector machines (SVMs), and the characteristic that Bayesian theory can make use of sample information and prior knowledge to simplify prediction model and optimize parameters．Firstly, the mechanism of BTP is analyzed, the Bayesian theory and LS-SVM are elaborated respectively, and the Bayesian evidence framework is applied to least squared support vector machine(LS-SVM） regression in order to infer non-linear models for predicting a time series．On the first level of inference, model parameters are selected and on the second level the hyper-parameters are selected．The kernel parameter are tuned on the third level framework，and on this level the relevant inputs are selected．A LS-SVM model is proposed on the basis of the Bayesian LS-SVM models. The results reveal that the BTP of sinter can be accurately predicted by this model even with small samples and poor information. It is concluded that the LS-SVM model is effective with the advantages of high precision, less samples required and simple calculation.