The study examines the association of social background, academic-related and institution-related factors with indicators of higher education dropout risk, i.e. students' perceived probability of graduating and their consideration of leaving their studies. Bivariate analysis and multilevel logistic models were used to analyse data from 1533 students from 25 study programs of the University of Zagreb, Croatia. The results confirmed the assumption that higher education dropout risk may be associated with different factors (i.e. social background, academic and institution-related characteristics), but also that there may be differences regarding the two indicators of perceived dropout risk. Moreover, the analysis revealed that perceived dropout risk may be viewed as part of a process of self-selection in which a combination of different factors leads a student to withdraw from higher education. Accordingly, the impact of the covariates of dropout risk should not be viewed only relative to each other, but should be evaluated in the context of educational decision-making net of academic ability.