There is widespread public concern at the high drop-out rates among apprentices in Australia. This paper argues that this concern may be misplaced as it fails to allow for eventual completion over time. We argue that a methodology that concentrates on a single training episode only is less appropriate to evaluating training outcomes in a regime where apprentices routinely record multiple episodes of training. This paper uses a multi-state approach to model the behaviour of a large sample of apprentices over a 60 month period which proves well suited to the episodic nature of cases observed in the data.