Torque converters are durable fluid couplings that can provide output torque multiplication. Blade leaning angle represents the angular position of a blade chord with respect to its radial reference line, and it is an important blade variable regarding both hydrodynamic performance and manufacturability of a torque converter. In traditional design processes, blade leaning angles are often determined based on experiences of engineers; hence, this study proposed a design approach using the combination of computational fluid dynamics (CFD) and optimization. Two CFD models were developed to design blade leaning angles. A steady-state periodic CFD model was employed for the parameter study and the optimization, and a transient full three-dimensional (3D) model was performed to study the flow mechanism and evaluate the performance with higher accuracy. Design of experiment (DOE) technique was employed to investigate the relationship between blade leaning angles and hydrodynamic performance, and a reduced cubic model was derived from the results. It was found that blade leaning angles had profound effects on torque converter performance; a large blade leaning angle intensified the flow blockage effect, thus resulting in a lower mass flowrate and torque capacity. Seven torque converters with different blade leaning angles were tested to validate the obtained numerical results, and the test data were found to be in good agreement with the CFD predictions. Finally, the hydrodynamic performance of the base model torque converter was optimized by a multi-objective genetic algorithm.