In this paper, Grondelaers and Van Hout’s statement that “the highest nonvirtual stratum of Belgian Dutch is documented by the speech of Belgian teachers” (2011, p. 219) is put to the test. Using production data from two corpora of contemporary spoken Dutch, mixed models binary logistic regression was carried out, focussing on 11 phonological and morpho-syntactic variables. The results show that teachers indeed use significantly less non-standard variants than other highly educated professionals. Moreover, there is also a difference between teachers of Dutch and teachers of other school subjects. By means of an exploratory content analysis of the sociolinguistic interviews, a few possible explanatory factors are discussed: teacher training, hypocorrection, and a difference in linguistic expectations. Especially the latter factor seems to be in play: teachers feel more pressured to adhere to the standard norm than the informants from the ‘laymen’ corpus.