This is a qualitative research that draws on Fairclough’s Critical Discourse Analysis methodology to analyze the discourse practices of the mathematics teacher educators in initial teacher training colleges in Malawi. The study involved four mathematics teacher educators in two teacher training colleges located in two different regions of Malawi. Specifically the study explored the following questions: 1) What are the discourse practices that mathematics teacher educators display in their descriptions of multilingual mathematics classrooms? 2) a) What are the discourse practices that mathematics teacher educators display in a college mathematics classroom? b) How do they make available the discourse practices for the student teachers to draw on? Data was collected through pre-observation interviews, classroom observations, reflective interviews and focus group discussions with the mathematics teacher educators. This study has shown that while there are some disconnections between the discourse practices produced in a school multilingual mathematics classroom and a college mathematics classroom, some of the discourse practices that mathematics teachers produced in a college mathematics classroom reinforces the common discourse practices being produced in multilingual mathematics classroom. There are three common discourse practices that were displayed in a college mathematics classroom. These discourse practices are: Initial-Response-Evaluation (Pimm, 1987), traditional lecturing and group discussions. I observed that the IRE and traditional lecturing discourse practices were accompanied by directive discourses for procedural control, and the procedural discourse was the prevalent discourse in all the discourse practices produced. iv Three major themes have emerged from the data analysis. Firstly, the research findings indicate that the mathematics teacher educators regard multilingualism and the language practices that come with it such as code-switching more as a problem rather than a resource for teaching and learning. Secondly, code-switching in college mathematics classroom is not as spontaneous as is research shows it to be in schools; rather it is very much controlled and restricted. Thirdly, the dilemmas of code-switching as discussed by Adler (1998, 2001) are more acute in teacher training colleges, mainly because of the mismatch in the Language-in-Education Policy (LiEP) in schools and tertiary level.