Dynamic mode decomposition (DMD) has been applied to the wake of a NREL-5MW wind turbine, to characterize the most dynamically relevant coherent structures characterizing this flow. The decomposition has been applied on a snapshot dataset obtained by Large-Eddy Simulation of the flow impinging on the wind turbine, whose tower and nacelle are modeled by the immersed boundary method, whereas rotor blades are modeled using the actuator line method. The Sparsity-Promoting DMD algorithm allows one to select a limited number of dynamic modes optimally reconstructing the snapshot sequence. Among the largest-amplitude selected modes, we found the tip vortices, oscillating at an angular frequency equal to three times the rotational frequency of the turbine. Interestingly, the remaining selected modes are characterized by low frequencies and large-scale spatial structures, reaching the frequency range of the wake meandering. This small set of dynamic modes is highly relevant for the formulation of reduced-order models.