While of the rectangular keep there remain many, and some very perfect, examples both in England and Normandy, the SHELL KEEP, though once the most common of the two, has rarely been preserved, and is seldom, if ever, found in a perfect or unaltered condition. There is a difference of opinion as to the date of the introduction of these keeps, whether a little before or a little after the other type. The shell keep, being invariably connected with early earthworks, might be supposed to be the older form; and Arundel, the only castle mentioned in Domesday as existing in the time of the Confessor, has a shell keep; but a tolerably close examination has failed to discover, either at Arundel or elsewhere in England or in Normandy, any masonry of very early character, probably none that can safely be attributed to the eleventh century. The fact seems to be that the early timber structures, which are known to have been erected originally on the moated mounds, were found to be very defensible, and so were retained by the Norman lords until they were able to replace the timber by masonry. The rectangular keeps were either on new sites, or on sites not defended by very strong earthworks, so that their construction, from the first, was in masonry, and thus it came to pass that the shells of masonry, though always connected with the older sites, were of later date than the solid towers. Even at Durham, a castle recorded to have been built by the Conqueror, and of which the keep must always have been on the present mound, though the chapel and connected buildings may be his work, the shell keep contains no Norman masonry; and if, as is to be supposed, there was once a Norman shell, it was probably the work of one of the Conqueror’s sons, or even of Henry II. A shell keep is always placed upon a mound, either natural or artificial. Of those on natural hills, the most considerable are Belvoir, Durham, and Lewes, but the masonry of the two former is not original, though built upon the old lines. Dunster, the Tor of the early Lords Mohun, has been examined, and the only trace of its keep is a fragment of a drain. Montacute, where the hill is wholly natural, has also been cleared of masonry. But by far the larger number of these mounds are either wholly artificial, or of a mixed character. In plan and dimensions these keeps are roughly governed by the figure of the mound on which they stand. Most are polygons of ten or twelve sides, not always equal. Some are circular, others polygonal outside and circular within. Others are slightly oval, others more complex in plan. York, for example, approaches to a quatrefoil, 64 feet by 45 feet, and this seems to have been the plan of the long-since destroyed keep of Warwick. Their diameter is rarely less than 30 feet, and seldom exceeds 100 feet.