To investigate the origin of the high strength of thin films, in-situ cross-sectional TEM deformation experiments have been performed on several metallic films attached to rigid substrates. Thermal cycles, comparable to those performed using laser reflectometry, were applied to thin foils inside the TEM and dislocation motion was recorded dynamically on video. These observations can be directly compared to the current models of dislocation hardening in thin films. As expected, the role of interfaces is crucial, but, depending on their nature, they can attract or repel dislocations. When the film/interface holds off dislocations, experimental values of film stress match those predicted by the Nix-Freund model. In contrast, the attracting case leads to higher stresses that are not explained by this model. Two possible hardening scenarios are explored here. The first one assumes that the dislocation/interface attraction reduces dislocation mobility and thus increases the yield stress of the film. The second one focuses on the lack of dislocation nucleation processes in the case of attracting interfaces, even though a few sources have been observed in-situ.