This paper examines the existence of social learning in agriculture in Ethiopia. We use a ‘random matching within sample’ technique to collect data on social networks and elicit details of the relationships and information exchange between network members, complementing the analysis with information on self-reported networks. We find that, while kinship or membership in certain groups, informal forms of insurance, or having frequent meetings with network members are all associated with a higher probability of forming an information link, none of these are correlated with observed innovative behavior such as the adoption of row-planting. This may suggest that behavior is more likely to be affected by the nature of information that passes through the network, rather than the number of information links. In support of this, we find that information links that exclusively involve discussions on farming or business matters are indeed associated with a higher likelihood of adopting row-planting. We use econometric strategies to isolate social learning from that of correlated and contextual effects. After controlling for factors that might otherwise generate spurious correlation, we find a strong evidence of network externalities in the adoption of row-planting techniques and also in farm productivity. Our results imply that extension services and other programs that promote agricultural innovations and seek yield improvement may benefit from social networks but they may be more effective if they identify the ‘right’ networks, that is, the ones that exclusively involve information exchange regarding agriculture. This further implies that investment in group formation, rather than simply using existing networks, may be a beneficial strategy.