Finding relevant video content is important for producers of television news, documentanes and commercials. As digital video collections become more widely available, content-based video retrieval tools will likely grow in importance for an even wider group of users. In this thesis we investigate language modelling approaches, that have been the focus of recent attention within the text information retrieval community, for the video search task. Language models are smoothed discrete generative probability distributions generally of text and provide a neat information retrieval formalism that we believe is equally applicable to traditional visual features as to text. We propose to model colour, edge and texture histogrambased features directly with discrete language models and this approach is compatible with further traditional visual feature representations. We provide a comprehensive and robust empirical study of smoothing methods, hierarchical semantic and physical structures, and fusion methods for this language modelling approach to video retrieval. The advantage of our approach is that it provides a consistent, effective and relatively efficient model for video retrieval.