Deep intronic FGF14 repeat expansions have been identified as a frequent genetic cause of late-onset cerebellar ataxias, explaining up to 30% of patients. Interruptions between repeats have previously been identified to impact the penetrance in other repeat expansion disorders. Repeat interruptions within FGF14 have yet to be characterized in detail. We utilized long-range PCR, Sanger sequencing, repeat-primed PCR, Nanopore, and PacBio sequencing to distinguish the repeat motifs, mosaicism, and number of repeat interruptions present in FGF14-related ataxia patients and unaffected individuals. We identified 28 individuals with an expanded repeat length (≥250 repeats) in the FGF14 gene after a previous screening of 367 patients with late-onset ataxia. Additionally, a cohort of 192 unaffecteds was screened for repeat expansions in the FGF14 gene where we found 12 expansion carriers. We applied advanced genetic methods to investigate the repeat motif. In total, the 40 individuals had expansions ranging from 232 to 486 repeats (SD=60) and 20 had repeat interruptions, including complex motifs such as GAG, GAAGGA, GAAGAAAGAA, GAAAAGAAGAAGGAAGAAGGAA, GAAAAGAAGAAGGAA, and GCAGAAGAAGAAGAA. We calculated the longest pure GAA length from the long-read data for all 40 individuals. When comparing the pure GAA tract between patients and unaffecteds, clusters were apparent based on greater or less than 200 repeats. Five ataxia patients with interruptions still had a remaining pure GAA expansion >200. We observed an association of the pure GAA length with age at onset (p=0.012, R2=0.263,). Unaffected individuals had a longer interruption length compared to the patients (p=0.010). Mosaic divergent repeat interruptions were discovered that affect motif length and sequence (mDRILS), which varied in number and mosaicism (frequency: 0.37-0.93). The mDRILS correlated with pure GAA length (p=0.022, R2=0.334), with a higher mosaic frequency of interruptions in unaffecteds compared to patients (unaffecteds: 0.90; patients: 0.67; p=0.010). We demonstrate that i) long-read sequencing is required to detect complex repeat interruptions accurately; ii) repeat interruptions in FGF14 are mosaic, have various lengths, and start positions in the repeat tract and can thereby be annotated as mDRILS, which iii) enabled us to establish a categorization based on remaining pure GAA repeats quantifying the impact of mDRILS on pathogenicity or age at onset, dependent on the interruption length and position, with high accuracy. Finally, we iv) provide evidence that mosaicism stabilizes pure GAA repeats in interrupted FGF14 repeat expansions.