Molecular diffusion is one of the most important factors to be considered when predicting oil recovery by application of gas injection. In this paper, a fluid characterization methodology including the estimation of molecular diffusion coefficients, which directly express the effect of molecular diffusion under reservoir conditions, is presented. In addition, the impact of diffusion on oil recovery by considering gas injection to a tight fractured gas condensate / heavy oil reservoir is also examined. We estimated the Fickian molecular diffusion coefficients from a combination of experimental data using a PVT cell and simulation matching based on coefficients calculated by the Leahy-Dios and Firoozabadi (2007) equation. As a result, we successfully estimated the coefficients and found that they largely deviated from the anticipated values. By an extension of a published study (Imai et al., 2012), using actual field oil and methane under different pressure conditions, we found that diffusion coefficients were smaller under higher pressure conditions. This implies that the higher the density and viscosity of the oil, the smaller the coefficients. Moreover, the diffusion processes in heavy oil systems were almost 10 times slower than those in light oil systems. Considering a tight fractured gas condensate reservoir and a heterogeneous heavy oil reservoir, a flow case study with gas injection accounting for the tuned coefficients was numerically simulated. The simulation study showed that the fractured case with molecular diffusion resulted in an oil recovery improvement of almost 10% as compared to the case with no molecular diffusion. The improvement was, moreover, larger than that in the case without experimentally tuned coefficients. This result implies that accurate estimation of the molecular diffusion coefficients under reservoir conditions is crucial for estimation of oil recovery with gas injection.