Heavy oil recovery by VAPEX appears to be a promising IOR technique as it uses less energy than SAGD and, if CO2 is injected, can also provide a means of disposing of excess CO2 in the subsurface. Nonetheless field application of this process has been limited due to concerns that favourable laboratory recoveries may not scale up to the field. In particular previous laboratory studies of VAPEX in porous media have obtained significantly higher production rates than predicted either by analytic models derived from Hele-Shaw experiments or numerical simulations. The discrepancy between experiment and models has been explained by assuming greater mixing between vapour and oil than would be expected from molecular diffusion. Justifications for this increase include convective dispersion, an increased surface area due to the formation of oil films on sand grains, imbibition of oil into those films and a greater dependence on drainage height. Convective dispersion seems to be the most plausible mechanism. This paper investigates the role of convective dispersion on oil recovery by VAPEX using a combination of well characterized laboratory experiments and numerical simulation. A first contact miscible fluid system was used so that all mechanisms contributing to increased-mixing apart from convective dispersion were eliminated. Longitudinal and transverse dispersion coefficients were measured experimentally as a function of flow-rate and viscosity ratio. Vapex drainage experiments were then performed over a range of injection rates. The laboratory measurements of oil drainage rate were compared with those predicted by the Butler-Mokrys analytical model and numerical simulation using either molecular diffusion or convective dispersion. Using measured convective dispersion improved prediction of oil drainage rate by 50%. The numerical model was then used to investigate the impact of rate (through viscous to gravity ratio and Peclet number), well separation and reservoir geometry on recovery.