Studies of social networks in organizations confront analytical challenges posed by the multilevel effects of hierarchical relations between organizational subunits on the presence or absence of informal network relations among organizational members. Conventional multilevel models may be usefully adopted to control for generic forms of non-independence between tie variables defined at multiple levels of analysis. Such models, however, are unable to identify the specific multilevel dependence mechanisms generating the observed network data. This is the basic difference between multilevel analysis of networks and the analysis of multilevel networks. The aim of this article is to show how recently derived multilevel exponential random graph models (MERGMs) may be specified and estimated to address the problems posed by the analysis of multilevel networks in organizations. We illustrate our methodological proposal using data on hierarchical subordination and informal communication relations between top managers in a multiunit industrial group. We discuss the implications of our results in the broader context of current theories of organizations as connected multilevel systems.