The Scale-Adaptive Simulation (SAS) approach has emerged as an improved unsteady Reynolds-Averaged Navier-Stokes (URANS) formulation to bridge the gap between the less accurate commonly used URANS and the computationally expensive hybrid RANS/LES for highly separated unsteady flows, e.g. dynamic stall. However, while the SAS has been successfully used at several occasions, it has not yet been tested for the complex case of dynamic stall. Therefore, the present study analyzes the SAS predictions of dynamic stall on a vertical axis wind turbine at a chord Reynolds number of 5 x 10(4) and a reduced frequency of 0.125. The analysis is based on comparison of the SAS predictions of the blade aerodynamics and the turbine power performance against the corresponding URANS and hybrid RANS/LES predictions. The results show that the SAS predictions are closer to hybrid RANS/LES than URANS with respect to: (i) the instant of the bursting of the laminar separation bubble (LSB), the leading-edge suction collapse, the formation of the dynamic stall vortex (DSV) and the trailing-edge vortex (TEV) and the shedding of the TEV; (ii) the size and strength of the TEV; (iii) the DSV-TEV interaction; (iv) the drag prediction during the downstroke. On the other hand, both URANS and SAS fail to corroborate with hybrid RANS/LES with respect to: (i) the instant of the formation of the LSB and the shedding of the DSV (the stall angle); (ii) the drag jump at the stall angle; (iii) the lift values during the downstroke; and (iv) the chordwise extent of the LSB.