Young highly educated workers developed in the 70's and 80's a preference for working in larger cities. As a consequence highly educated young workers in 1990 were over-represented in cities, in spite of the lower wage premium they earned for working in crowded metropolitan areas if compared to their older colleagues. This can be an equilibrium only if young workers enjoy some benefits in cities and are willing to pay for them. In our model, the extra-benefit of working in cities is given by a dynamic externality of human capital. Agglomerations of educated workers arise endogenously, as workers are attracted to dense areas, which improve their learning from others. If the skills accumulated in cities are easily transferable, it is efficient for educated people to work in dense areas while they are young and move to less dense areas when they become mature workers. Once the learning period is over, workers are attracted to smaller and less dense locations where there is less competition from other skilled workers and housing price is lower. Our model explains why young workers were attracted into large cities in the 70's and 80's: this was the era of increased flexibility, of the success of versatility rather than specificity of skills. Small firms thrived, and therefore the transferability of skills increased. The model also gives an account of why, once they accumulated their human capital, some of the workers moved to smaller towns.