Introduction The construction of a brain network from neuro- biological data is really a critical point. It is usual- ly represented in terms of a graph, consisting of nodes and links between pairs of nodes, called edges of the graph. In large-scale analysis, nodes relates to brain regions, while links represent sta- tistical correlations between such regions. Differ- ently, small-scale analysis generally focuses on a specific neural system, either a sub-system of in- terest (retina, cortical column, etc.) or a small but entire nervous system (e.g., Caenorhabditis ele- gans, fruit fly, or grasshopper). Generally, the number of data obtained perform- ing any kind of experiment (M/EEG, fMRI, PET, etc.) is huge, for example 100 nodes could lead to nearly 10,000 data. As a consequence, some threshold must be introduced, in order to empha- size the main interesting outcomes. The choice of suitable thresholds is really important since these should be able to separate meaningful links from weak and non-significant links that might repre- sent spurious connections, particularly in func- tional or effective networks. Archives Italiennes de Biologie, 154: 78-101, 2016. DOI 10.12871/00039829201625 Corresponding author: Paolo Finotelli, Department of Mathematics F. Brioschi, Politecnico di Milano, Piazza Leonardo da Vinci 32, I-20133 Milano - Email: paolo.finotelli@polimi.it A Statistical Proposal for Selecting a Data-depending Threshold in Neurobiology P. FINOTELLI 1 , P. DULIO 1 , G. VAROTTO 2 , F. ROTONDI 2 , F. PANZICA 2 1 Department of Mathematics “F. Brioschi”, Politecnico di Milano, Milan, Italy; 2 Neurophysiology and Diagnostic Epileptology Operative Unit, “C. Besta” Neurological Institute IRCCS Foundation, Milan, Italy ABSTRACT In this paper we propose a new methodology for introducing thresholds in the analysis of neuro- biological databases. Often, in Neuroscience, absolute thresholds are adopted. This is done by cutting the data below (or above) predetermined values of the involved parameters, without an analysis of the distribution of the collected data concerning the phenomenon under investigation. Despite an absolute threshold could be rigorously defined in terms of physic parameters, it can be influenced by many different subjective aspects, including cognitive processes, and individual adaptation to the external stimuli. A possible related risk is that, mainly in experiments also de-pending on personal reactions, a significant portion of meaningful data, relevant for that specific task, could be neglected. In order to reduce these deviations, we are proposing to adopt a task-dependent approach, based on the comparison between the collected data and some database concerning a different task, assumed as a baseline. After giving the necessary theoretical back-ground, we test our methodology on real EEG data involving two subjects in a musical task. In addition to some natural results, new and unexpected neurological links can be emphasized and discussed. Key words Brain networks • Functional connectivity • Graphs • EEG data • Musical task • Threshold