Space-based observations of solar-like oscillators on the red giant branch have led to detections of oscillations in thousands of stars. Fitting stellar models to these data, however, is more difficult than in main sequence stars. This is partly because of how surface effects (systematic frequency differences caused by poor modelling of the near-surface layers) potentially lead to misidentification of the non-radial modes. I present a method to circumvent this problem and results of applying this method to three red giants in double-lined eclipsing binary. We obtain consistent results for the magnitude of the surface effect in all three stars, which agree with existing predictions from three-dimensional hydrodynamic simulations. We also find that the best-fitting stellar models are usually significantly larger and more massive than determined by the orbital solutions.