The development of polylactide (PLA)/cellulose nanocrystals (CNC) nanocomposites in the aim to increase gas barrier properties has been widely studied, but gave rise to conflicting results. To better understand the underlying gas transport mechanisms, PLA/CNC nanocomposites were produced at concentrations between 5 and 50 wt%. The role of the PLA/CNC interface for the permeability was investigated using three different CNC surface grafts, namely lauric acid (slightly interacting with PLA), stearic acid (non-interacting with PLA), and poly glycidyl methacrylate bearing a reactive epoxy end-group (crosslinking with PLA). While PLA/CNC composites contained a large number of aggregates, homogenously dispersed nanocomposites were obtained with surface modified CNCs even at high concentration of CNC (up to 30 wt%). All nanocomposites had better O2 and CO2 barrier properties than PLA. Surprisingly, the nanocomposites prepared with surface grafted CNC did not perform better than the ones prepared with neat CNC. The diffusion coefficient was successfully modeled with Nielsen's law, showing the importance of the shape factor and resulting tortuosity for the barrier performance notwithstanding PLA/CNC surface compatibility. The analysis of the dual-mode O2 sorption isotherm showed the creation of a similar quantity of supplementary sorption sites in nanocomposites prepared with neat or lauric acid grafted CNC. PGMA grafts induced a higher number of sorption sites. The most important advantage of CNC surface grafts was the shielding of CNC against water vapor uptake. As a result, the O2 barrier properties could be maintained constant up to 90% relative humidity.