In this paper, I study the behavioural motivations underlying job referral decisions. In a field experiment in Addis Ababa, Ethiopia, I show that when choosing who to refer for a job, individuals trade off personal benefits and altruistic considerations, with important implications for the efficiency and equity of the referral process. Using complete data on urban social networks and generating a panel of real work and referral opportunities over multiple sessions, I first show that workers rely on reciprocity. This leads to both significant on-the-job productivity losses and persistence in the exclusion of less connected individuals. This dynamic reciprocity is reduced under incentivised referrals, where workers are paid according to the output of their referral, which makes them screen more productive work-ers. Second, I find that peripheral workers use job referrals strategically to enlarge their network: they are more likely to establish new and reciprocated links, with connections persisting after 18 months. I show that these findings are consistent with a network-based job referral model where individuals trade off social payoffs and altruistic considerations. My findings suggest that conventional job referrals through social networks can reinforce labour market inequalities and prevent less socially connected individuals from getting ac-cess to jobs. However, when given referral opportunities, these individuals can manage to escape exclusion even in the long-run. Policy-makers could exploit this and provide sub-sidised temporary jobs as linking opportunities, with the goal of alleviating long-term youth unemployment.