Abstract The successful testing of a gravity drainage process, using the horizontal wells in the Underground Test Facility (UTF) in Alberta, Canada indicates that high recovery and economical oil-steam ratios are achievable. However, several major technical challenges must be resolved to extend this successful pilot experience to commercial operations and to the many different heavy oil and extra-heavy oil reservoirs. This paper provides a consistent package of experimental data on the development of gravity drainage using horizontal wellbores exposed to a variety of injection-production strategies. Critical parameters that control process performance such as initialization time, completion, reservoir permeability and initial oil mobility are highlighted. A 60×21×3 m section of a homogeneous field was scaled using Pujol and Boberg's criteria to a 2-D visualization cell (60×21×3 cm). One minute of experimental time represents 7 days of field time. The used scaling criteria offered adequate scaling of gravitational forces, fluid properties were preserved in the lab and the same initial conditions in the field and lab. Results demonstrate improvement of process performance by using different and novel injection strategies. Introduction A theory for the application of steam assisted gravity drainage (SAGD) in the recovery of extra heavy oil was developed by Butler and McNab and Butler and Stephen. In the original theory the major assumptions were that steam chamber starts over the entire vertical height of the reservoir and along the length of the horizontal well. Butler later introduced modifications to the theory using a TANDRAIN assumption where a tangent was drawn from the production point to the steam chamber interface. In addition, the modification took into account the rise of the steam chamber. The theory provides a useful tool for rapid, and approximate, assessment of oil drainage from homogeneous reservoirs. However, to describe the entire process (both initialization and drainage) and address the non-uniform distribution of steam along the well-bore as well as reservoir heterogeneities, other methods need to be developed. Field Application of The Gravity Drainage. The UTF phase A project, Edmunds, was the first field demonstration of the SAGD process. Three horizontal well pairs were used (in each pair, an injector was located 5 m above the producer). Conduction heating in the space between the two wells was chosen as the method for initialization of gravity. Steam was circulated in the tubing and out of the annulus. In the observation wells located just above the injection well, the temperature decreased along the length of the well and suggested that the far end of the well was probably not very hot. Initial breakthrough between the wells will generally occur in a localized region. Proper initialization procedures are required to bring the entire length of a well pair into active drainage. In addition, optimization of production control is essential for successful application of the process. Too much drawdown between the wells will produce large quantities of steam. For a small drawdown, liquid will continue to build-up between the wells. This will result in the injector being "drowned" and the height of the chamber decreased (impede drainage). The drowning mechanism is poorly understood at this stage. Heterogeneous Reservoirs. Joshi reported results on using SAGD with vertical and horizontal injectors. He found that vertical injectors with a horizontal producer gave faster recovery than using a horizontal injector/horizontal producer in reservoirs with shale barriers. He also indicated that vertical fractures perpendicular to a horizontal injector improved oil recovery rate as compared to a horizontal injector/horizontal producer. Yang and Butler studied two types of reservoir heterogeneities. First, reservoirs containing thin shale layers and secondly, reservoirs containing layers of different permeabilities (two layers reservoir). They found that a short horizontal barrier did not greatly affect the general performance of the SAGD process. P. 77^