In this thesis, the challenges associated with the automatic prediction of 3D shape and pose of subjects from both RGB and RGB-D images will be discussed, and the unique difficulties that arise when these subjects are animals. The motivation for my attempt to address some of these challenges and the reasons why the area of the automatic extraction of animal pose is a topic of interest will be presented. The contributions of this thesis will then be outlined. A brief history of animal motion capture and pose prediction will be given, leading on to a discussion of the most recent methods and their results. This will lead on to the work carried out during the undertaking of this thesis. Details will be given of experiments involving the creation of statistical pose models and how these can be leveraged when predicting the pose of a horse and dog. Next, the process of collecting a multi-modal dataset of canine shape and pose will be discussed and examples given of the more expressive canine pose models this data can be used to produce. In addition, a canine shape model is created. A potential use of this large dataset involves the prediction of canine shape and pose from RGB-D images. This comprises of a pipeline where a neural network first predicts the 3D joints of the animal, and a pose-prior model is fit to these locations in order to produce skeleton joint rotations. From this skeleton, a posed mesh can be created. The pipeline is compared to other current methods and quantitative evaluations are given. Finally, I will present my conclusions and discuss possible future directions for this research.