Framing is one of the most central, applicable, and contested theories in communication research. At the heart of the debate on framing is the question of operationalizing and measuring emphasis frames. We harness novel computational tools to propose a new method for inductive identification of frames. We argue and demonstrate that frame elements could be identified using topic modeling, and that frame elements can then be automatically grouped into frame “packages” using community detection techniques applied to the topic network. Building upon recent conceptual and methodological developments in framing research, we introduce a new approach, the Analysis of Topic Model Networks (ANTMN). We demonstrate the applicability of our method in case studies where framing theory is developed and fairly consistent, and in exploratory ones where it is not, using three diverse U.S. news corpora: the coverage of political candidates in Senate races (n = 8,337 articles), foreign nations (n = 18,216), and infectious diseases and epidemics (n = 5,005). We conclude by discussing the theoretical, methodological, and practical implications of ANTMN.