This paper presents an analysis of the stability and accuracy of different Lattice Boltzmann schemes when employed for direct numerical simulations of turbulent flows. The Single-Relaxation-Time scheme of Bhatnagar, Gross and Krook (BGK), the Multi-Relaxation-Time scheme (MRT) and the Regularized Lattice Boltzmann scheme (RLB) are considered. The stability and accuracy properties of these schemes are investigated by computing three-dimensional Taylor-Green vortices representing homogeneous isotropic turbulent flows. Varying Reynolds numbers and grid resolutions were considered. As expected, the BGK scheme requires sufficiently high grid resolutions for stable and accurate simulations. Surprisingly, the MRT scheme when used without any turbulence model fails to obtain mesh convergence for the type of flow considered here. The RLB scheme allows for stable simulations but exhibits a strong dissipative behavior. A similar behavior was found when employing the mentioned LBM schemes for numerical simulations of turbulent channel flows at varying Reynolds numbers and resolutions. The obtained insights on accuracy and stability of the considered Lattice Boltzmann methods can become useful especially for the design of effective turbulence models to be used for high Reynolds number flows.