Due to the water-coning problem, cycle steam stimulation (CSS) in the heavy oil reservoirs with bottom water is often less effective, and the oil recovery is even below 10%. Steam assisted gravity drainage (SAGD) is the oil-producing process with a constant pressure-drop (about 0.30 MPa), and it is a potential technique for this reservoirs. Through the implementation of SAGD, bottom water could be effectively controlled. Aiming at the LD heavy oil block in Bohai offshore oilfield, the SAGD performance in heterogeneous heavy oil reservoir with bottom water was numerically studied in this paper. In these simulation models, the water was broken into three components (connate water, injected water and bottom water) to study the water producing in SAGD process. Thus, the influences of startup approach, oillayer thickness, water thickness and the distance between well-pair and bottom-water on the water rising were all simulated. Thereafter, a set of numerical simulations were performed to assess the shale issues in SAGD process, e.g. the vertical and horizontal position of shalebarriers, the shale distribution range, the barrier permeable condition and the macroscopic vertical permeability.Results indicated that bottom water reduced the ultimate recovery of SAGD process by about 10% of the OOIP. The startup by steam-circulation was much suitable for the bottom water heavy oil reservoir instead of CSS approach. The bottom water tremendously reduced the startup pressure-decline rate, and thus the startup-time was prolonged. The distance between well pair and water zone had a great influence on the SAGD performance, and a small distance would delay the beginning time of the steam-chamber rising. For reservoir heterogeneity, the vertical and horizontal heterogeneity have great influence on the drainage process, especially the shale cases. It tremendously decreased the recovery rate by about one time. This investigation could be used as a tool for the successful design of SAGD process in heavy oil reservoirs with bottom water.