The speed and intensity of the appearance changes that occur during the formation of facial expressions provide important information about the underlying meaning of the expression itself. In the past we have demonstrated the effectiveness of using locally linear embedding with facial shape information for estimating the dynamics of facial expression. This approach was only suitable for specific expressions, where the appearance change was principally due to a movement or distortion of the shape of facial features. However, for some facial expressions, the variation in the shape of the facial features is very subtle. These expressions are mainly characterised by the variation in the texture of the face. Hence such expressions are not amenable to the previous approach. In order to estimate the dynamics of these types of expressions it is necessary to develop nonlinear appearance models that incorporate texture information. In this paper we use LLE to estimate the manifold of texture variation due to facial expression. We show that the resulting manifold effectively captures the underlying dynamics of facial expression and that it provides a suitable representation for differentiation between posed and spontaneous expressions.