Nowadays internationalisation has become an aggressive agenda of growing strategic importance to tertiary education around the globe, driven by the influence of a highly competitive economy domestically and globally. A large and growing body of literature has suggested that English Medium Instruction (EMI) is the foremost instrument to advance universities in rankings systems in numerous universities in non-Anglophone countries in order to compete for elevated positions in an international arena. Thailand, among these countries, now enthusiastically uses EMI programmes in various academic disciplines. Each university has different mechanisms for EMI implementation. The context of this study is a regional university which is in an elite group of Thai public universities. The aim of this study is to pay close attention to two language mechanisms in this university. That is, the English Linguistic Gears are an in-house innovation in encouraging content lecturers to use English in one of three levels (‘Gears’): Gear One requires 25 percent of English of the class time; Gear Two, 50 percent; and Gear Three, 75 percent. EMI lecturers are expected to weave these gears into a Language Pillar, an overseas approach derived from Content and Language Integrated Learning (CLIL). The language and subject content are given equal weight in CLIL and it is assumed that the second language is used more often than the first, and that the teachers are competent to teach both the academic content and second language skills. At classroom level, the empirical literature on language practices under these mechanisms in this context is limited. It is not clear what factors inform and shape the lecturers’ language beliefs and practices. More importantly, much uncertainty still exists about the relationship between language practices in EMI classrooms. This thesis is the first substantial qualitative case study in Thailand to investigate the lecturers’ language beliefs and practices in EMI classrooms. Data were collected from August 2016 to January 2017 from six lecturers in courses in humanities and social sciences through semi-structured interviews, classroom observations, stimulated recall interviews, focus group discussions, documents, and researcher reflections. Grounded Theory was adopted to analyse all data. The findings showed that the lecturers were faced with the dilemma of language practice between two mechanisms. That is, all lecturers selected Gear One, which was directed at using 25 percent of English and allowed 75 percent of Thai. In contrast, the Language Pillar expected the lecturers to use English to the full, alongside explicitly teaching the English language. The lecturers’ language beliefs revealed that they were aware that insufficient English skills of both lecturers and students would negatively impact on lecturers’ instruction and students’ understanding. Their language practices showed that the lecturers emphasised the need not only for the students to understand the academic content but also to promote rapport with the students. In so doing, they used considerably more Thai than English, although both languages had specific roles to play. In addition, the lecturers’ reflection on their language practices revealed that external factors (policy, classroom infrastructure, and students) and internal factors (the lecturers’ own language preferences and proficiencies) were the crucial factors that shaped and informed their current language beliefs and language practices. The findings presented that code-switching and translanguaging were the lecturers’ way out of the dilemma in English medium instruction classrooms. One possible explanation for these findings was that language had its own flexibility in using (including abilities to use, methods for use, and purposes of use) and fluidity of movement. The major finding of the study is that a continuum emerged when the use of code-switching for classroom social interactions flowed into translanguaging for instructional functions. To understand the growing phenomenon of EMI in the research site, a comprehensive and multi-dimensional conceptual framework, ROADMAPPING (Dafouz & Smit, 2016, 2020) was applied to provide a holistic view of the language practices in and beyond classrooms. This framework identifies six relevant components, Roles of English (in relation to other languages) (RO), Academic Disciplines (AD), (language) Management (M), Agents (A), Practices and Processes (PP), and Internationalisation and Glocalisation (ING), and each component intersects with and impact on the others. The discussion of the findings in this study pays close attention to all components of the framework in order to explain crucial interrelationships shaping EMI in the local context. The research has yielded implications on contextual, practical, and theoretical areas. Thus, it has thrown new light on language practices in tertiary education EMI programmes for future dual language policy, practice, and training for related agencies such as institutional policy-makers, content lecturers, teacher-trainers, researchers, and beyond.