This work aims at assessing the impact of heterogeneities on heavy oil recovery through SAGD process by using numerical models of a reservoir analogue of meander belt. These models are obtained with different scales of upscaling of the geological model.Meander belts consist of point bar deposits, characterized by a 3D complex internal architecture, with different scales of heterogeneities, which distribution is associated with the depositional processes.Based on a 3D outcrop description of a meander belt analogue to the Canadian heavy-oil fields, the approach includes four steps: 1/ the construction of a reference static reservoir model based on a very fine description of the outcrops in terms of architecture and geological heterogeneities, 2/ upscaling of the grid at different scales using different methods in order to evaluate their impact on the heterogeneity distribution in the reservoir, 3/ reservoir SAGD simulations using horizontal well doublet (steam injector and producer) across the meander belt, to assess the impact of upscaling of heterogeneities effect and on heavy oil production.The impact of heterogeneities on simulation results are evaluated for several upscaling stages. Results show that heterogeneity distribution has an impact on fluid flow at different stages of production. On the fine gridded model, small scale heterogeneities impact the steam chamber development and fluid flow in the wellbore vicinity at the beginning of the steam injection, whereas large scale heterogeneities strongly influence oil recovery during the whole recovery process and lower the efficiency of the reservoir drainage. On coarser grids, the effect of small scale heterogeneities can be smoothed, depending on the upscaling stage. The geomechanical effect is not taken into account in this work, the goal being to assess the impact of heterogeneities on oil recovery.The performance of SAGD is clearly linked to the steam chamber development, which depends on the degree of heterogeneities present in the reservoir. The simulation workflow and the sensitivity study based on different upscaling methods contribute to better restore the heterogeneity distribution in the reservoir. The negative impact of these heterogeneities during the oil recovery must thus be quantified in order to monitor the thermal production at crucial periods of production.