Pupil behaviours inform about cognitive load and performance during online activities (Laeng et al., 2012). Contrary to other ocular responses such as Gazing, which heavily relies on cortical networks, pupil behaviour is thought to rely on neurotransmitter systems located in the brainstem (Alnaes et al., 2014). We have recently shown that Gazing can inform about underpinnings of memory binding deficits in patients with Alzheimer's disease (Fernández et al., 2018). As a cortical response, Gazing is more prompted to cognitive bias whereas pupil behaviour, might provide an unbiased biological measure of cognitive performance during a task known to be a marker for AD (the short-term memory binding task – STMBT). This study investigated if the combined analysis of pupil behaviour during the STMBT test could enhance the sensitivity to detect AD. Data were collected from 13 controls and 13 age and education matched patients with AD. Patients were diagnosed based on DSM-IV criteria. We used EyeLink 1000 eyetracker to collect oculomotor data during STMB performance. The STMBT assesses the ability to recognise changes in two consecutive arrays of bicolored objects which had to be remember as wholes (bound colours - BC) or as parts (unbound colours - UC). We compared behavioural and ocular data across Groups, Task Conditions, and Memory Stages. Mean pupil behaviour significantly increased in controls from UC to BC condition a change not observed in AD patients (t=1.97). Controls' and AD patients' pupil behaviour did not differ during the UC condition (t=0.63) but it did during the BC condition (t=-5.19). ROC analyses revealed that pupil behaviour during performance on the BC condition achieved 100% sensitivity and specificity (See Figure 1). Mean pupil behaviour across conditions of the VSTMBT, memory stages, and Groups. Binding colours into bicolored objects is known to be a cognitive demanding task. Such demands are taxed by pupil behaviour only in controls, thus indicating that the mechanisms responsible for extracting such a cost are impaired by AD neurodegeneration and that was true for 100% of our patients. A few AD makers have achieved such a classification power. We suggest that combining VSTMB performance and pupil behaviour would offer biomarker evidence of AD pathology.