Maximising the power output of wind farms is vital in order to reduce reliance on fossil fuels for generating energy in a timely manner. However, wind turbines often operate in the wake of other turbines leading to a reduction in the wind speed and the resulting power output. In a bid to increase the total farm power output, wake steering can be applied to redirect the wakes and minimise their effect on nearby turbines. This optimisation problem has typically been addressed using analytical wake models due to their computational efficiency. This allows many different configurations to be tested when searching for an optimum. However, as the wake models are missing important physical features, they may find an optimum that does not reflect the true optimum of the real wind farm. Large eddy simulation (LES) allows for non-linear and unsteady fluid phenomena to be included and a solution found that is closer to the true optimum. Bayesian optimisation (BO) presents an efficient optimisation strategy for expensive to evaluate black box functions and is therefore a good candidate for use with LES. In the current work, a multi-fidelity (MF) BO strategy is implemented to exploit the use of cheap approximations in the form of analytical wake models, whilst also benefiting from the additional physics captured by LES. The MF-BO is implemented by constructing a MF surrogate model using the nonlinear autoregressive gaussian process (NARGP) approach of [1], the structure of which is displayed in figure 1(a). This MF surrogate model is used to determine the configuration and fidelity of each experiment. The surrogate models and the location of each experiment in the parameter space is shown for a three turbine yaw optimisation in figure 1(b). An improvement in the total power output of the wind farm is found, using a limited number of LES evaluations, as shown in figure 1(c). This method will be used for finding wake steering stratagies for larger wind farm layouts.