This paper describes a method by which a computer can autonomously acquire training data for learning to recognize a user's face. The computer, in this method, actively seeks out opportunities to acquire informative face examples. Using the principles of co-training, it combines a face detector trained on a single input image with tracking to extract face examples for learning. Our results show that this method extracts well-localized, diverse face examples from video after being introduced to the user through only one input image. In addition to requiring very little human intervention, a second significant benefit to this method is that it doesn't rely on a statistical classifier trained on a pre-existing face database for face detection. Because it doesn't require pre-training, this method has built-in robustness for situations where the application conditions differ from the conditions under which training data were acquired.