Abstract: To control the roof subsidence in gob,the longwall panel is generally backfilled after mining. The rock burst risk evaluation method of the island backfill longwall panel (IBLW) is different from that of the normal longwall panel due to the complex distribution of its abutment pressure. In order to provide a theoretical basis for the stability analysis of surrounding rock during the retrieval of IBLW,by establishing a “filling body-island pan-el-roof ” model based on the evolution of the overburden structure,the evaluation method of abutment pressure in front of the working face and the criterion for rock burst were obtained. Research results are as follows:① The subsidence of the overburden strata above IBLW and the horizontal associated movement of the overburden strata above the gobs on both sides were deter- mined by the filling effect which was also a key factor of overburden structure evolution. ② Influenced by roadway tun- neling and the de-stressing of large-diameter boreholes,the obvious regional difference of bearing capability of IBLW was observed. The load transferred from the overburden strata was mainly born by the elastic area of IBLW. The stabili- ty of IBLW was affected by its boundary conditions and layout of the roadways. ③ The filling materials deformed due to the subsidence movement of the roof,which stopped when mechanical equilibrium was reached. The stability of IBLW was mainly controlled by the compres-sive deformation and elastic modulus of the filling material. ④ The rock burst risk of IBLW was dependent on the static abutment pressure in the elastic area and subsidence of the roof above the gob. To prevent this kind of rock burst,the static abutment pressure can be reduced by injecting water into the coal seam,hydrofracturing and op-timizing the layout of longwall panels. Also,the roof subsidence can be controlled by im- proving the filling rate of the gob,optimizing the consolidation and compressive properties. This research result provides a technical guidance for rock burst prediction and control.