This paper examines organizational learning through a multilevel network lens. We assess how inter-personal knowledge transfer is sustained by the organizational structure of interunit work-flow ties andby the level of specialism of the connected units.To do this, we apply Multilevel Exponential Random Graph Models on data collected in a multiunitgovernment institution in Italy.Results indicate that our approach allows simplifying and better understanding of organizationallearning. Units are more likely to retain knowledge transfer ties within their boundaries. Unit boundary-spanning tends to occur only when knowledge transfer ties are sustained by hierarchical interunitwork-flow ties.