In this paper we present the evolution of a bidirectional leader which appeared in ‘virgin’ air. In the observations we did not find any delay between the onsets of the two leader ends. The positive leader end presented a single channel, brighter and slower than the negative end. On the other side, the negative end was branched, less bright and twice as fast as the positive end. The attachment of one of the negative channels with a pre-existing intra-cloud channel showed that the polarity reversal point in the bidirectional leader remained fixed. Recoil leaders appeared along old negative channels of the bidirectional leader. Recoil activity started after the first branching of the positive leader. The discussion is aimed at improving current bidirectional leader models.