The single relaxation time (SRT) and the revised matrix (RM) lattice Boltzmann models are compared for simulations of three dimensional forced isotropic turbulence with resolutions of 1283 and 2563, respectively. Guo et al.[1] forcing technique with the same forcing parameters and conditions is used to investigate the ability and stability of the two models for simulations of forced turbulence. Some new aspects and results have been confirmed such as the superiority of the MRT model to simulate forced turbulence in comparison with the SRT model. To overcome the stability shortage problem of the lattice Boltzmann models, it is proposed to apply the Courant-Friedichs-Lewy condition (CFL) [2] by multiplying the velocity input with the CFL coefficient which is less than unity and by dividing the output velocity data by the same CFL coeffi-cent. The initial velocity field is chosen as u(x, 0) = 0 and the force is injected at low wave-numbers with a fixed forcing amplitude of 10−4 for all cases. The single relaxation time is set to τ = 0.503 in all SRT simulations. Results show that the obtained turbulent velocity fields yield universal characteristics as proven in previous theoretical, experimental and numerical studies. The Taylor Reynolds number for the simulations are found as : Rλ(SRT) = 62 and Rλ(M RT) = 65 for the resolution of 1283 and Rλ(SRT) = 107 and Rλ(M RT) = 82 for the case of 2563, respectively. To test the weak incompressibility for the SRT model in comparison to the MRT model, the probability distribution function (PDF) of the density ρ is depicted and it is found that ρ is almost unity at all timesteps for the MRT case, while a clear disturbance about unity is observed for the SRT case. Time variation’s statistics such as the Taylor Reynolds number Rλ, the Taylor microscale λ, the Kolmogrov microscale η and energy spectra are calculated and depicted for all cases. One of the disadvantages of the MRT case is the simulations time which is found about twice the SRT simulations time, where about 80 hours are needed for simulating 12000 timesteps in the 2563 case in comparison to 42 hours for the SRT model using the same hardware resources.