This paper examines the use of individual, dyadic and system-level analyses in the study of relational data in organizational networks. We argue that dyadic analyses are particularly appropriate when the dependent variable is quantitative and/or involves multiple behaviors. We show that system-level analyses, by aggregating potentially signi?cant information, provide a less grounded account of the relations across networks than do dyadic analyses. Using examples from a study of corporate political behavior, we contrast dyadic analyses with those at both the individual and system-levels. Variables measured in raw dyadic form consistently perform better in accounting for similarity of corporate political behavior than do variables measured by taking system-level properties into account. Our ?ndings suggest that although individual and system-level analyses are useful in a number of situations, dyadic analyses are a ?exible means to examine the effects of multiple networks at multiple levels.