Relying on the analysis of computer mouse movements in constrained two-alternative forced choice tasks (mouse-tracking) allows to study the fined grained dynamics of decision processes. When adopting a situated social cognition perspective, it allows studying the influence of sensorimotor, individual and social components, and has been used to study lexical decisions, social perception, attitudes, or stereotypes. Widespread mouse-tracking software programs produce data composed of trajectories, thus combining spatial and temporal information for each trial. In most published papers, these trajectories are reduced to summary statistics such as the Maximal Deviation (MD) or the Area Under the Curve (AUC). These summary statistics are then analyzed through classical means, such as ANOVA, or more recently linear mixed models. In some cases, and to avoid losing too much information while simplifying the data, the dimension of the whole trajectory is reduced by aggregating data points in time bins, or through principal component analysis to obtain a set of meaningful components. While mouse movements are either sampled at a fixed rate, or directly generated from the sequence of mouse-related events, they end up being represented as a sequence of points (with x, y, and time coordinates), allowing the application of the aforementioned methods. Nevertheless, such data is fundamentally functional from a statistical perspective, reflecting the continuous movement of the human hand controlling the mouse. An alternative way to analyze this data is thus to rely on growth curve analysis, but when considering the non-linearity inherent to mouse trajectories, this approach is far from trivial to apply in practice. We instead propose to project trajectories on a functional basis (here arbitrarily made of B-splines). Once the basis parameters are fixed, this provides a unique decomposition of each trajectory, made of components representing the local influence (in space and time) of factors of interest. We thus simplify the original object (of infinite dimension) while keeping its spatio-temporal continuity, producing (multi-dimensional) data that can be analyzed by standard statistical methods such that ANOVA or mixed effects models. We illustrate this approach on two datasets – one with a fixed sampling rate, collected in a community sample with the task focusing on gender stereotypes, the other event-based, conducted among young children categorizing racially morphed faces – showing that the method is robust to many sources of variability and statistical problems, which cannot be easily dealt with using the classical methods.