Introduction) Indexing is a strategic activity in Library and Information Science, playing a central role in the organization and retrieval of knowledge within Information Retrieval Systems (IRS). In this context, relevance emerges as a key criterion for guiding subject analysis, especially given the cognitive, situational, affective, and pragmatic complexity of information-seeking processes. (Objective) This article aims to propose the Inferential Model of Subject Analysis guided by Relevance (MIAAR), articulating epistemic, heuristic, and interpretive foundations to enhance thematic representation in IRS. (Methodology) The study adopts a qualitative approach grounded in a critical review of specialized literature, focusing on the multidimensional nature of relevance, the quality of indexing, and the inferential dimensions of subject analysis. (Results) The model proposed comprises five interdependent stages: contextual reading, conceptual selection, situational evaluation, epistemic pertinence, and representational formulation. MIAAR aims to qualify the mediation between documentary content, knowledge organization systems, and users’ judgment criteria, fostering more coherent, context-aware, and responsive thematic representations. (Conclusions) The model contributes to enhancing the communicational effectiveness of IRS by aligning thematic representation with users’ informational needs and interpretative strategies. Future studies are encouraged to empirically validate the model and to explore its potential in supporting the development of controlled vocabularies and ontologies.