This Glossop Lecture is about landslides and their slip surfaces at residual strength in clays. Particularly in southern England, but also elsewhere, landslides in infrastructure cuttings and many natural slopes are commonly found to be slowly moving compound landslides with a component of their basal shear surfaces following a particular bed (or ‘slide-prone horizon’). A selection of both historical and modern case records of this type of landslide are presented briefly. The geotechnical conditions that give rise to this occurrence are discussed, and the dominant factor relates to the dip of the strata, which must be of low inclination for the landslide mass to remain in place over the critical clay bed in the geological sequence after sliding has been initiated. Observations of the slip surfaces in the field lead to the conclusion that the bedding-controlled elements of this type of landslide develop along thin, slide-prone or slide-susceptible, horizons in the bedding. The question of what caused the formation of those horizons in the first place is answered by putting forward two hypotheses to explain why bedding-controlled slip surfaces form where they do, and considering the evidence for or against each of them. The conclusion is reached that despite the attractiveness of the concept that these slip surfaces form by a progressive failure mechanism at the junction of two materials with dissimilar properties, the alternative concept that they occur where there is a bed of slightly enhanced smectite content better fits the observations. The mechanisms for such local changes in clay mineralogy are linked to inputs of volcanic ash at the time of deposition. Definitive proof of concept is, however, lacking, but taking into account how clay sediments are deposited in sedimentary basins, this paper makes suggestions for future lines of enquiry. Even now, nearly a half-century after Skempton’s seminal Rankine Lecture that introduced the concept of residual strength of clays to the wider geotechnical profession, the corpus of data is rather limited. Some of the datasets are shown to exhibit remarkable similarities, and the implications of this tend to support the preferred explanation of the origin of slide-prone horizons.