In neuronal axons, almost all microtubules are oriented with their growing end (+end) away from the cell body (they are oriented ‘+end out’). Molecular motor proteins rely on this orientation to efficiently move vital cargo to the distal regions of the axon. Despite 30 years of research, the mechanism that establishes this unique microtubule configuration, and thus long-range transport, remains unknown. In my thesis, I developed several machine learning algorithms for microscopy image analysis that exploit novel developments in deep learning. The first tool I developed, KymoButler, enabled me to automatically analyse microtubule growth behaviour in Drosophila melanogaster axons with unmatched precision. I discovered that +end out microtubules grow longer than -end out microtubules. Furthermore, I found evidence that the microtubule anti-catastrophe factor, p150, is partially responsible for this difference in growth times through a molecular gradient in the distal tip region of the axon. Together with a mathematical model of microtubule growth, these observations implied dramatic differences in average microtubule length, whereby -end out microtubules were short and unstable, and +end out microtubules grew to hundreds of microns in length. Accordingly, both pharmaceutical treatment with the microtubule destabilising drug Nocodazole and removal of p150 decreased +end out microtubule growth times and the fraction of +end out microtubules in axons. These findings suggested a simple mechanism that organises axonal microtubules. Since -end out microtubules remain labile they are more likely to depolymerise completely and thus disappear from the axon. This would result in a larger proportion of long +end out microtubules in the axon. My results pave the way towards a deeper understanding of how microtubules orient during development to support molecular transport, potentially shedding new light on many neurological pathologies that are characterized by transport deficiencies.