The Euphrates, gathered into a single channel, flows very grandly through the narrow gorge. At first the hills slope down almost to the water’s edge, but afterwards they draw back and leave room for a tract of level ground by the stream. An hour and a half from Wâdî Mâliḥ the valley widens still more, and on the opposite bank the great castle of Ḥalebîyeh lifts its walls from the river almost to the summit of the hill, a towered triangle of which the apex is the citadel that dominates all the defile (Fig. 46).[53] Twenty minutes lower down, the Mesopotamian bank is crowned by the sister fortress of Zelebîyeh. It is a much less important building. The walls, set with rectangular towers, enclose three sides of an oblong court; the fourth side--that towards the river--must also have been walled, and it is probable that the castle approached more nearly to a square than at present appears, for the current has undermined the precipitous bank and the western part of the fortifications has fallen away. The masonry is of large blocks of stone, faced on the interior and on the exterior of the walls, while the core is mainly of rubble and mortar. There are six towers, including the corner bastions, in the length of the east wall, and between the two central towers is an arched gate. On the north and south sides there is now but one tower beyond the corner. Each tower contains a small rectangular chamber approached by an arched doorway. The court is covered with ruins, and on either side of the gate there is a deep arched recess. Under the north side of the castle hill there are foundations of buildings in hewn stone, but the area of these ruins is not large. The name Zelebîyeh carries with it the memory of an older title; in the heyday of Palmyrene prosperity a fortress called after Zenobia guarded the trade route from her capital into Persia, and all authorities are agreed that the fortress of Zenobia described by Procopius is identical with Ḥalebîyeh. Procopius states further that Justinian, who rebuilt Zenobia and Circesium, refortified the next castle to Circesium, which he calls Annouca. The Arab geographers make mention of a small town, Khânûḥah, midway between Ḳarḳîsîyâ (Circesium) and Raḳḳah,[54] and the probable identity of Annouca and Khânûḳah has already been observed by Moritz.[55] But I think it likely that the flourishing mediæval Arab town was situated not in the confined valley below Zelebîyeh but at Abu ’Atîḳ, where the ruin field is much larger. It may be that there was a yet older settlement at Abu ’Atîḳ, and that the stone foundations there belonged to the town of Annouca which stood at the head of the defile, while the castle of the same name guarded the lower end.