It has been repeatedly shown that mismatch negativity (MMN), an event related potential measurement, reveals differences between dyslexic children and age-matched controls. MMN reflects the automatic detection of deviance between a stream of incoming sounds presented to the passive listener, and deficits in MMN (i.e. attenuated amplitudes) have been especially reported in dyslexia for detecting differences between speech sounds (e.g./ba/vs./da/). We performed an association analysis in 200 dyslexic children. This analysis focused on two MMN components, an early MMN (188-300 ms) and a late MMN (300-710 ms), and the dyslexia candidate genes KIAA0319 and DCDC2 on chromosome 6. Additionally, we imputed rare variants located in this region based on the 1000 genomes project. We identified four rare variants that were significantly associated with the late MMN component. For three of these variants, which were in high LD to each other, genotyping confirmed the association signal. Our results point to an association between late MMN and rare variants in a candidate gene region for dyslexia.