Background: Fluency is both a diagnostic criterion and a clinical label inherited by the field of dementia from cerebrovascular aphasia. As the multidimensional entity it is it has no clear working definition for diagnosis since its different components can be traced back to distinct functional networks and brain areas. For instance, the former nonfluent variant of Primary Progressive Aphasia (PPA) now subsumes the agrammatic/nonfluent (PPA-nf) and the logopenic (PPA-L) subtypes. While the former is in need of redefinition after the discovery of Primary Progressive Apraxia and the persisting difficulties in describing frank agrammatism, the latter has been questioned due to its pathological link to Alzheimer’s Dementia (AD), a condition that also exhibits linguistic deficits (Leyton & Ballard, 2016). Mild Cognitive Impairment (MCI) - especially its multifocal type- also displays linguistic impairments and has been linked to dementia, especially the temporomesial type with AD. Given its clinical heterogeneity the identification of differential profiles could be key for the early diagnosis of specific dementia aetiologies. However, speech production and fluency measures seem to offer low test-retest reliability in these populations, especially in short speech samples (Ahmed, Haigh, de Jager & Garrard, 2013), raising doubts about their diagnostic and descriptive value even in the face of perceived disfluency and/or incipient communicative impairment. Procedure: Samples of picture-based oral narratives by Spanish-speaking participants were transcribed using multi-layer annotation accounting for different language levels with an ad-hoc system for disfluency labelling and pause distribution reporting. The cohort included 13 mildly-impaired AD patients, 15 amnestic-temporomesial(aMCIa) and 24 amnestic-executive (aMCIb) MCI subjects, as well as 4 PPA-nf and 10 PPA-L patients in the mild to moderate stages, and 11 healthy elderly controls (HC). The occurrence of false starts, repetitions, repaired sequences, abandoned structures, automatisms, perseverations and echolalia as well as intra-clausal and intra-phrasal pauses was evaluated and compared to other linguistic measures in order to find differential performance patterns. Results: Compared to HC and aMCIb subjects APP-nf participants exhibited the classic disfluent pattern with a significant disfluency rate, particularly in repetition and false start counts, combined with a greater number of intra-clausal pauses and less correct words. APP-L individuals showed a disfluent pattern with repaired sequences, lexical and semantic paraphasias and phonological errors, in addition to an also significantly high intra-clausal pause rate. Participants with AD displayed a longer total silence time and more pauses, alongside a significant number of lexico-semantic paraphasias. aMCIa patients performed similarly but lexical and semantic errors were the only statistically significant parameters. Conclusion: Disfluency markers proved more useful than classic measures for describing group performance and their correlations with linguistic indexes (fig.2) indeed seem to reveal differential (dis)fluency patterns that can be related to specific functional circuits. aMCIa subjects show incipient linguistic deficits in an almost comparable rate and distribution to those of AD participants, which could be pre-dementia signs related to early memory loss. This calls for a closer look into lexico-semantic and speech planning deficits as they could be linked to temporo-parietal damage in the early stages of pre-clinical dementia.