The document proposes a reversible data hiding technique for video frames based on a Matched Neighbour Predictor. It uses three consecutive video frames, where the middle frame is used as a reference to predict pixels in the first and third frames. Prediction errors between the original and predicted frames are calculated and used to embed data by expanding the errors. Experimental results on two test videos show the proposed technique embeds data at a rate of bits per pixel with a peak signal-to-noise ratio maintained. The technique allows exact recovery of both the original video frames and embedded data.