Smart water injection or low-salinity water flooding is an enhanced oil recovery technique for fractured reservoirs. The mechanism of low-salinity water flooding is performed by altering the wettability of rock towards water-wetness. This study presents an evaluation of low-salinity water flooding based on a brown field fractured reservoir properties by using explicit simulation of spontaneous imbibition at a fine scale (single and multiple matrix block level), and assessing the validity of dual-medium simulation models, which will be further used to perform modelling at the full field scale. Some reservoir parameters were varied within plausible range to quantify their impact on the recovery mechanism. To mitigate the biases of dual-medium models, upscaling and pseudoisation techniques were applied to match the behaviour of explicit fine-scale single-porosity model. A particular emphasis on the modelling of salt diffusion is made through one-dimension model. It was found that, whereas pressure diffusion is rather well represented by the usual matrix-fracture exchange coefficient formalism, the underlying assumption of pseudo-steady state seems to be inappropriate for salt diffusion phenomenon. When salt diffusion is the main driver for low salinity water imbibition, for example at the early times of a switch from high salinity to low salinity, it appears that the salt diffusion process occurs in a transient regime that cannot be properly represented by the pseudo-steady state regime assumption. This has an impact on the modelled incremental recovery and consequently on the low salinity flooding efficiency evaluation.