The maximum extent of a glaciation is often confused with the coldest part of a glaciation. During the Last Glaciation, the date of the Last Ice Maximum Extent (LIME) is diachronous very early at high latitudes close to the traditional date in the intertropical mountains. The classical Last Glacial Maximum (LGM) never corresponds to the LIME with the exception of intertropical regions. At middle and high latitudes advances of the LGM only correspond to surges of a lately thickened ice sheet. The same structure also exists at the level of the entire Cenozoic of a whole glaciation or of much shorter events as the scale of a whole Bond’s cycle. For each type of event the reactional sequence is autocyclic: a rapid warming ends a slowly cooling trend due to the higher susceptibility of lower latitudes to external forcing (orbital and solar activity). Moisture supply from an ice-free ocean is the sine qua non to build glaciers. The syngenetic extent of the cold desert toward the equator is the limiting factor for ice volume: when glaciations develop in intertropical mountains with the development of sea ice in the polar zones, glaciers recede by precipitation starvation with restricted sedimentation. During a glacial era in parallel with ice cap building, the progressive sea level lowering and the subsequent enhanced aridity counterbalance the generalization of a world-scale glaciation. This concept could be also valid for the Neoproterozoic Glacial era and should probably invalidate the snowball hypothesis.