This paper is concerned with assessing the predictability for a blast furnace (BF) system through nonlinear time series analysis of silicon content in blast furnace hot metal, in three representative blast furnaces with different volumes. The results indicate that the predictability of silicon content in hot metal of the selected BFs approaches that of a totally predictable system while it departs from that of a totally unpredictable system greatly. The predictability of silicon sequences conversely renders a strong indication of the presence of a deterministic mechanism ruling the dynamics of the ironmalcing blast furnace process. In the meantime, the results also provide important information on selecting the minimum number of macroscopic variables and the delay time needed for reconstructing the dynamics of these blast furnace systems. Moreover, the above information is found to be independent of the size of these blast furnaces. However, there is a slight difference in the complexity of the silicon sequence collected from different types of blast furnaces. As one might expect, larger difficulty will be encountered in predicting the silicon content of a pint-sized blast furnace because of the presence of the larger complexity.