We have developed single- and multiple-kernel appearance-based models for human faces that allow descriptive parameters to be estimated from portraits and other images of faces. Although the training of the model has significant novel features, including a new covariance regularisation method, the estimation phase is a form of Conditional Density Estimation (CDE) -- a classical tool of probability modelling. The estimated parameters include gender, age, ethnicity, pose, and facial expression. In the seminar I will review applications of face description, discuss performance of existing automated systems, identify key issues and give a live demo. I will also show how the CDE mechanism allows modification of portraits along age and other dimensions.