This paper investigates the existence of peer effects in academic outcomes by exploringspecificities in the student's admission process of a Brazilian federal university, which works as a naturalexperiment. Individuals who are comparable in terms of previous academic achievement end up havingclassmates with better or worse performance in college because of the assignment rule of students toclassrooms. Thus, our identification strategy for estimating peer effects on academic outcomes eliminates theendogenous self-selection into groups that would otherwise undermine the causal inference of peer effects.Overall, our findings showed that joining a class with high-ability students damages academic achievementsof the lowest-ability students at UFMG. Although male and female students are both negatively affected bybeing in the first (better) class, we found gender differences. Specifically, being at the bottom of the betterclass make females take less radical decisions compared to male students in the sense that female studentscontinue to study even though with lower performance (reduced GPA and credits earned) while male studentsseem to be more prone towards dropping out (increased number of subjects – or even University registration– cancelled and reduced attendance in classroom). We have also found other heterogeneities in peer effectsin college in terms of class shift, period of admission, area of study and parents’ education. This study is anecessary step before investigating the impact of peer quality on after-graduating decisions using the samenatural experiment. This will allow us to deepen our understanding of how peer effects can also have long-lasting impacts.