This contribution focuses on the matching of local features between images. Given a set of query descriptors and a database of candidate descriptors, the goal is to decide which ones should be matched. This is a crucial issue, since the matching procedure is often a preliminary step for object detection, scene identification or image matching. In practice, this matching step is often reduced to a specific threshold on the Euclidean distance to the nearest neighbor. We first introduce a robust distance between descriptors, making use of the Earth Mover's Distance (EMD). We then propose an a contrario framework for the matching procedure, which enables us to control the number of false alarms. This approach yields validation thresholds automatically adapted to the complexity of the descriptor to be matched and to the diversity and size of the database. The method makes it possible to detect multiple occurrences and to rate the validated matches according to their meaningfulness.