Abstract The association between plants and flower visitors has been historically proposed as a main factor driving the evolutionary change of both flower and pollinator phenotypes. The considerable diversity in floral morphology within the tribe Antirrhineae has been traditionally related to pollinator types. We used empirical data on the flower visitors from 59 Antirrhineae taxa from the literature and our own field surveys, which provide an opportunity to test whether flower phenotypes are reliable predictors of visitors and pollinator niches. The degree of adjustment between eight key floral traits and actual visitors was explored by testing the predictive value of inferred pollinator syndromes (i.e. suites of floral traits that characterise groups of plant species related to pollination). Actual visitors and inferred pollinator niches (categorisation of visitors’ association using a modularity algorithm) were also explored using Linear Discriminant Analysis (LDA). The bee pollinator niche is correctly classified for flowers with dull corolla colour, without nectar guides, as the most important predictor. Both predictive value and statistical classification prove useful in classifying Antirrhineae taxa and the bee pollinator niche, mostly as a consequence of the high proportion of genera and taxa with occluded corollas primarily visited by bees. Our predictive approach rendered a high Positive Predictive Value (PPV) of floral traits in the diagnosis of visitors/pollinator niches. In particular, a high PPV was found for bees as both visitors and forming pollinator niches. In addition, LDA showed that four pollinator niches are well defined based on floral traits. The large number of species visited by bees irrespective of pollinator syndromes leads us to hypothesise their generalist pollinator role, despite the phenotypically specialised flowers of Antirrhineae.