Even though it has been recognized that prospective teachers' conceptions of the nature of mathematics, and of mathematics learning and teaching shape their teaching decisions, and, thereby, students' engagement and achievement, to date no research has examined these conceptions from a person-centered perspective taking into account year of teacher education program. Cluster analysis revealed three distinct profiles: anti-socioconstructivist, socioconstructivist, and flexible. Although the same driving dimensions were present at the different stages of the training, the characterization of these three profiles fluctuated, in tandem with the process of construction of a professional identity, the increase of the field experience and the discovery of the complexities of the profession. The analysis of the pedagogical practices support the non-linearity of the beliefs-practices relationship. Plain Language Summary: This contribution attempts to shed light on pre-service secondary school teachers' beliefs about mathematics, and mathematics teaching and learning and the relationships between their beliefs and their pedagogical practices. To achieve our main objective, the beliefs data were collected by questionnaire from 646 pre-service secondary school teachers in different years of their teacher education program. In order to take into account inter-individual differences in pre-service teachers' beliefs, the data were submitted to cluster analysis. Cluster analysis is a technique that seeks to discern structure in a set of data by grouping respondents according to the similarity of their responses. Cluster analysis revealed three distinct profiles: anti-socioconstructivist/pro-traditional, socioconstructivist/anti-traditional, and flexible. Although the same driving dimensions were present at the different stages of the training, the characterization of these three profiles fluctuated, in tandem with the process of construction of a professional identity, the increase of the field experience and the discovery of the complexities of the profession. Our results provide new insights on three aspects. First, it appeared that he belief profiles of future mathematics teachers are more complex than the commonly accepted binary model. Not only is there the presence of an intermediate profile, but the other profiles are not that clear-cut. Second, the analysis of these profiles and of the pedagogical practices acknowledged by the future teachers in our sample support the non-linearity of the beliefs-practices relationship. Our findings are consistent with, and complementary to, research that has shown the role played by pre-service experiences in this relationship. Finally, our findings support a more balanced approach to the teaching and learning of mathematics, instead of a desire to be on either end of the continuum (traditional vs. socio-constructivist). This contribution provides interesting results for the field. At the end of their training, less than 40% of that year's cohort defended an anti-socio-constructivist conception of the nature of mathematics, its teaching and learning. More than 40% defended a socio-constructivist approach and approximately one-fifth of those in their final year adopted a flexible perspective. Further, our results highlight the importance to encourage students to hybridize their conceptions and practices and thus to develop more appropriate teaching practices. Several studies would be interesting to carry out to complete this contribution: longitudinal approach, nuanced with qualitative research, etc. [ABSTRACT FROM AUTHOR]