Basic oxygen furnace(BOF) steelmaking endpoint prediction model is very important for endpoint hit of endpoint carbon content and temperature. Mutual information calculation is used to select input variables. To distinguish the importance of input variables to output variables, the input variables are weighted and the values are established by using particle swarm optimization algorithm. Finally, two support vector machine models are built to predict BOF endpoint carbon content and temperature. Simulations are implemented by using practical production data from a 180t BOF. The results show that proper variable selection and weighted pretreatment can improve the precision of prediction models effectively.