The building of a separation-reaction network for heavy feedstock requires an understanding of the reactivity for subfractions. This paper proposes a study on the coking reactivity for two heavy petroleum samples, vacuum residua (VR) from China Liaohe and Venezuela Orinoco crude oils, and their supercritical fluid extraction fractionation (SFEF) subfractions. The properties of feedstocks and their SFEF series were analyzed, including density, molecular, weight, elemental content, Conradson carbon residue (CCR), SARA components (saturates, aromatics, resins and asphaltenes) and structure parameters by nuclear magnetic resonance (NMR). All the samples were subjected to a laboratory-scale batch reactor to investigate the coking reactivity. Coke yield increased as the SFEF subfraction became heavier; while the yields of naphtha, diesel, and gas oil all decreased. The results show that the classic linear correlation between coke yield and CCR are not appropriate for separated fractions of VR. Instead, a negative power function was observed. Furthermore, the coke yield shows a linear relationship with aromaticity for both SFEF series. To precisely estimate coking reactivity for feedstocks and subfractions, a new SARA component based prediction model was carried out, which reveals the contribution of each component