Link-prediction is an active research field within network theory, aiming at uncovering missing connections or predicting the emergence of future relationships from the observed network structure. Here we propose a novel method for link-prediction that relies on the entropy maximization procedure introduced in the framework of network reconstruction. Given set of constraints (e.g. degrees, strengths), this algorithm quantifies the likelihood that any two nodes are connected by a link, a feature allowing us to apply the method to solve the link imputation problem and to provide weight estimation. Remarkably, our procedure also outputs proper confidence intervals for the weight estimation of any (missing) connection, thus overcoming one of the main limitations of common link-prediction algorithms (i.e. being applicable to binary networks only).