In the basic oxygen steelmaking process, the endpoint temperature of the molten steel is one of the key factors whether the molten steel is qualified for tapping. Currently, it mainly relies on the experienced operators to control the endpoint temperature of the molten steel, and the prediction precision may vary among different operators. In order to realize the effectively end-point steel temperature prediction of the basic oxygen furnace as well as to meet the requirement of different sizes of the converter mouth, a new method based on the flame temperature measurement at the converter mouth was proposed in this paper. Firstly, a fiber-optic spectrometer system in the visible and near infrared spectral range was designed which can real-timely and effectively realize the collection of the flame radiation information at the converter mouth. Secondly, in consideration of the actual temperature of the flame and the distance between the converter and the designed system, an improved calibration method instead of the halogen lamp was adopted, and the two-color method was employed for the flame temperature measurement. Then a regression model based on the support vector machine was built with the flame temperature and several other parameters of the steel-making process as the input variables of the model. Verification experiment was carried out on 68 industrial data collected in the steel-making workshop. The results show that the prediction accuracy of this method is superior to the experienced operators, and close to the sub-lance based method. As a result, the proposed method can provide a feasible and effective solution to the end-point steel temperature prediction for those small-sized and medium-sized converters.