Normative circular and dimensional models are the dominant structures for the organization of vocational interests in the scientific literature. However, it is increasingly recognized that not all individuals' interest configurations can be adequately represented by normative models. Adopting a person-centered, multidimensional perspective on vocational interests, the current study seeks to identify distinct profiles of interests based on RIASEC data that integrate interest configurations that align with and deviate from normal circular and dimensional structures. We also test the replicability of the profile structure, examine the likelihood of STEM degree choice as a function of profile membership, and investigate core personality predictors of interest profile membership. Latent profile analyses revealed six profiles of vocational interests, representing distinct combinations of the RIASEC interests (i.e., social-dominant, disinterested, high realistic-dominant, investigative-dominant, ambivalent, and conventional-dominant), which replicated entirely across independent subsamples. Furthermore, the profiles differed on the likelihood of STEM degree choice, with the conventional-dominant profile evincing the highest probability of choice and the social-dominant profile evincing the lowest probability of choice. Finally, results revealed that the Big-Five personality traits were differentially related to interest profile membership, largely in line with vocational interest theory. The present findings constitute novel evidence that a person-centered framework for the representation of interest configurations can accommodate both people's adherence to and deviations from normative structures for the organization of interests. The findings also underpin the use of all available interest information on individuals, rather than reliance on the two or three highest interest dimensions, to inform educational and vocational decision-making. •Latent profiles of vocational interests were identified.•The profiles replicated across subsamples.•Big-Five personality dimensions differentiated the profiles.•Profile membership was associated with the probability of STEM major choice.