The current active-latent paradigm of tuberculosis largely neglects the documented spectrum of disease. Inconsistency with regard to definitions, terminology, and diagnostic criteria for different tuberculosis states has limited the progress in research and product development that are needed to achieve tuberculosis elimination. We aimed to develop a new framework of classification for tuberculosis that accommodates key disease states but is sufficiently simple to support pragmatic research and implementation. Through an international Delphi exercise that involved 71 participants representing a wide range of disciplines, sectors, income settings, and geographies, consensus was reached on a set of conceptual states, related terminology, and research gaps. The International Consensus for Early TB (ICE-TB) framework distinguishes disease from infection by the presence of macroscopic pathology and defines two subclinical and two clinical tuberculosis states on the basis of reported symptoms or signs of tuberculosis, further differentiated by likely infectiousness. The presence of viable Mycobacterium tuberculosis and an associated host response are prerequisites for all states of infection and disease. Our framework provides a clear direction for tuberculosis research, which will, in time, improve tuberculosis clinical care and elimination policies.