We present a novel method for automatically synchronizing two video sequences of the same event. Unlike previously proposed methods, we do not put any restrictive constraints on the scene nor on the camera motions: our method can deal with independently moving cameras, wide baseline conditions, and general 3D scenes. It starts from five point correspondences throughout the video sequences, that are provided using wide baseline matching and tracking techniques. It is efficient, in that it can be implemented in a non-combinatorial way. The feasibility of the method is demonstrated by preliminary experimental results.