It is now timely to explore the challenges and opportunities of building life-long user models. These will capture salient aspects about the user over very long periods of time, from early childhood into old age. Life-long models will be critical if we are to realise the potential of personalisation that operates over the long term, as well as short-term personalisation that can be improved by long term data about the user. One of these benefits is a new form of augmented cognition. Another important class of personalisation will be in the context of improved life-long teaching and learning. This should take account of many elements that might be valuable as parts of a long term learner model. Notable among these is the learner's previous knowledge and its evolution, context-sensitive modelling of the user's response to learning activities and the learner's long term goals. To take just one other example, there is huge potential for radical improvements in the support for personal information management, with a long term user model enabling software tools to be far more effective than is currently possible with existing models in serving the different needs of each individual. While there is a considerable body of research into techniques for many aspects of user modelling, we now need to tackle the new challenges that need to be addressed for life-long user modelling. At a technical level, these include representational aspects which can take account of the effects of time, such as forgetting. Ontological issues will be fundamental since the life-long user model needs to be able to operate in relation to many different applications. On a very different level, there are critical challenges associated with privacy and user control which involve a combination of technical, social, user interface and other issues.