A revision of the data used to build the Caspian Sea level curve over the last 2200 years BP has been made based on a combination of geological and archaeo-historical data, using only those for which sufficient metadata were available. This compilation is completed by new sedimentological and palynological data from the south-east corner of the Caspian Sea, especially close to the known termini of the Sasanian Gorgan and Tammisheh Walls. A new calibration of the radiocarbon dates was used, i.e. with a freshwater offset reservoir of 351 ± 33 years. A literature survey of the Derbent lowstand indicated that this term has different definitions, depending on authors; it is thus to be used with caution. Here we therefore prefer to distinguish the mid-Sasanian lowstand and the later Medieval moderate lowstand. The “2600 years BP highstand” has not been found, mostly due to the calibration or recalibration of the datapoints used; data are indeed lacking at that time. Instead, a younger Parthian highstand (around 50 BCE–50 CE) is clearly defined. The maximal amplitude and speed of change of the Caspian Sea level were respectively of >15 m and 14 cm per year. Compared to last century, the latter rate is 25% higher, but the amplitude is more than five times larger. The climatic causes of the Caspian Sea level changes are discussed. It is far from a simple case of temperature forcing; temperature forcing may result in several effects, that may impact the Caspian Sea level variations in opposite ways. Moreover, human intervention on river diversion and natural hazards were likely, for several time periods.