An embedded Reynolds-Averaged Navier–Stokes blade element actuator disk model is used to investigate the performance of a closely spaced cross-stream fence of four turbines. The flow characteristics of such fences are found to be dependent on both the local turbine scale flow problem and the array in channel flow scale problem. The mean fence power is found to be less than that predicted for a single turbine with the same local blockage ratio (ratio of turbine swept area to surrounding flow passage area), but greater than that for a single turbine based on the global blockage ratio of the fence (ratio of total fence swept area to the cross-sectional area of the channel). Cross-fence variation in turbine performance is observed as a result of the differing resistance to bypass flow acceleration around the inboard and outboard turbines and depends on the operating condition of the turbines. Reducing turbine thrust, such as by changing the rotational speed of the turbine or by employing a pitch-to-feather power capping mechanism reduces turbine-turbine interactions and turbine performance becomes more uniform across the fence. An approximately 6% increase in the mean fence power can be achieved if a cross-fence differential blade pitch strategy is employed to maximise the lift to drag ratio along the majority of the blade span of each of the turbine blades.