Abstract This paper investigates three morphosyntactic alternations in Estonian – those between the exterior locative cases allative, adessive, and ablative and the corresponding postpositions peale ‘onto’, peal ‘on’, and pealt ‘off’. It is assumed that the influence of different predictors on speakers’ choices will be relatively stable in terms of the direction of those predictors, but the strength of these will vary. For each alternation, a random sample of the two outcomes (case vs. postposition) from the Estonian National Corpus is used, resulting in a total of 3,000 data points. Using properties of the landmark phrase as independent variables in mixed-effects logistic regression models, the choice of postpositions over case-marked realizations is predicted. The models fitted to the data confirm that the direction of the eight predictors investigated is the same across the alternations, with freqRatio, lemma, and mobility making the most significant contribution to the fit of all three models. The study further shows that the two alternating pairs that have a higher global frequency in Estonian (allative ∼ peale and adessive ∼ peal) behave in a similar way with respect to the predictors under study; the third, less frequent alternation (ablative ∼ pealt), differs from the other two in terms of the number and type of predictors that make a significant contribution to the model fit.