Investigation surges is important for studying patterns of water dynamics in the Northern Caspian. Modeling of wind-surges fluctuations in the Caspian Sea was performed using the model ADCIRC (ADvanced CIRCulation). For numerical experiments was constructed an irregular triangulation computational grid that consists of 71,523 elements with minimum cell size 500 m (off the coast) and maximum - 10 km (in the center of the sea). It was compiled relative to the absolute zero water level for the Caspian Sea equal to −28" role="presentation" style="position: relative;">-28-28-28 m BS (Baltic System), taking into account flooding and drying of the coastal land. The NCEP/CFSR reanalysis of surface wind at a height of 10 meters and the atmospheric pressure from 1979 to 2017 were used as the input data. Since the Northern Caspian is covered with ice in winter, the sea ice concentration of OSI-450 reprocessing were used. To assess the quality of the model, the obtained model level data were compared with level measurement on weather stations Tuleniy Island and Makhachkala from 2003 to 2017. The correlation coefficient for Tuleniy Island varied within 0.79-0.88, for Makhachkala - 0.67-0.79. The mean value of the root-mean-square error was about 0.11 m for Tuleniy, and 0.06 m for Makhachkala. If we exclude fluctuations in sea level with an amplitude of less than 10 cm, the correlation values increase on average by 0.04. Three main synoptic situations were identified that lead to the formation of surges of more than 1 meter on Tuleniy Island. The contribution of wind to the formation of surge height is predominant (92-100%) compared with the contribution of the horizontal pressure gradient (0-8%).