We present Geometric Data Analysis applied to a barometer data survey in order to construct the French “trust space”, define groups of French citizens as regards trust and study the evolution of trust over time. The data source stems from the survey entitled “Le baromètre de la confiance politique” (“Barometer of political trust”) carried out by the CEVIPOF (Center for political studies in Science Po, Paris). The data have been collected since 2009 on a yearly basis. The samples (about 1,500 persons) are designed to be representative of the French citizens registered to vote. A set of 22 questions that are relevant to 5 themes (trust in political and non-political institutions, economic institutions, interpersonal and personal trust) is selected for this study. The questions are bipolar (19 are coded on a Likert scale with 4 or 5-point and 3 are dichotomous). For constructing the space, we perform correspondence analysis (CA) after doubling the data and we weight the questions in order to balance the importance of themes. Individuals of Wave 1 (2009) are put as active individuals (reference wave), the ones of other waves are supplementary elements. Then, for studying the evolution of each question, we compute the weighted mean of the principal coordinates of individuals of each wave. For defining groups of individuals, we proceed to Euclidean clustering (AHC, Ward’s method) using the same distance between individuals than the one used in CA. The clustering is performed on the individuals of Wave 1; individuals of the other waves (2012, 2013 and 2014) are assigned to clusters. We propose a strategy for assigning supplementary objects to clusters stemming from an AHC. The assignment will be made downwards based upon the superior hierarchy at the local level of a node to one of its two successors until a cluster of the partition under study is reached. For each node, we define an assignment criterion based on the ratio of distances from the object-point to barycenters of both clusters that make up the node. The distances refer to the Mahalanobis distances associated with each cluster. This ranking rule enables us to take the shape of clusters into account. The methods are implemented in R that can be invoked from SPAD Software.