Peer effects are potentially important for optimally organising schools and neighbourhoods. In this paper, we examine how the gender of classmates and neighbours affect subsequent academic performance, the choice of academic tracks, university related outcomes, and the choice of university study. This is the first paper that: a) examines gender peer effects in school on longer term outcomes related to university and 2) examines gender peer effects using spatial variation and cluster analysis to construct the neighbourhoods. We employ two strategies. First we exploit within-school and -neighbourhood idiosyncratic variation in the proportion of females across consecutive cohorts in the twelfth grade. Using data for the universe of students in pubic schools in Greece between 2004 and 2009, we find that a higher share of females in school or neighbourhood improves females' and males' subsequent academic performance, increases the university admission rate for both genders, makes both genders more likely to enrol in university compared to education institutes and affects the choice of university study. In particular, we find that females are more likely to enrol in STEM related university departments when they have more females in school or neighbourhood. Next, we exploit the random assignment of students to classes within schools in the eleventh grade. To do this we use new data that we collected from 106 high schools. We control for potentially confounding unobserved characteristics of schools, classes and students that could be related to the proportion of females, and we exploit variation in the proportion of females across classes within the same school in a given year to obtain identification. We find that an 1% increase in eleventh grade female classmates improves girls subsequent academic performance in mathematics by 0.1 s.d and increases their likelihood to enrol in STEM related university departments by 1%.