Steel slags in different treatment processes (such as layer pouring slag from converter, rotating furnace slag from converter, casting slag, desulfurized slag, layer pouring slag from electric furnace and rotating furnace slag from electric furnace) were set as the research objects. Research on steel slag’s chemical composition and phase composition was carried out by X-Ray diffraction (XRD) and X-Ray fluorescence (XRF). The relationship of chemical composition, phase composition and activity was fitted by steel slag activity index prediction model with back-propagation neural network. The results showed that steel slag treatment process is different, its chemical composition and phase composition are obviously different. Steel slag activity index prediction model is of good compliance, and the relative errors are 2.42% and -2.54%, which can fully reflect the mapping relationship between input layer and output layer.