Cross-sectional geometry of a horizontal axis tidal stream turbine (HATST) blade was optimized using surrogate models and computational fluid dynamics (CFD) analysis. The blade thickness parameters of a 100 kW class HATST model, i.e., relative thickness and maximum relative thickness location, were varied to examine change of turbine performance in terms of power coefficient. Multiple surrogates such as response surface approximation, radial basis function, Kriging and weighted average surrogates were implemented to the CFD analysis results with design parameter variation to search the optimal design. It was found that the Kriging model was suitable for this HATST optimization problem as it produced the smallest cross-validation error and high accuracy. The optimized design enhanced the power coefficient by 17.9%, which shows a way to implement the present approach to tidal stream turbine design and optimization.