Multiple design variables modifications are carried out for a bidirectional flow turbine used in an oscillating water column wave energy converter to enhance its performance by maximizing the peak torque-coefficient (TC) and the corresponding efficiency (EFF), which are the objective functions of this problem. The Latin hypercube sampling technique selects samples from a designed space created by the design parameters defined for the blade sweep and aerofoil profile thickness modifications. The objective function values are obtained by solving Reynolds-averaged Navier-Stokes equations and are approximated by surrogate models. The models help in generating populations of the genetic algorithm, which finally produces a set of optimal designs in a Pareto optimal front. Only two extreme designs among the five clustered points are further evaluated by solving Reynolds-averaged Navier-Stokes equations to cross-check the validity of the optimization steps. It is found that the TC is increased by 33% and the EFF is decreased by 5% at one extreme cluster point, while the other extreme point gives that both the TC and the EFF are higher by 1.8% and 2.9%, respectively, as compared with the reference geometry. The optimal geometry has a wider operating range, which is an important parameter to get continuous power from a wave energy converter. Copyright (C) 2016 John Wiley & Sons, Ltd.