Spacers play a significant role in enhancing water permeation in membrane desalination processes but also increase pressure drop and specific energy consumption for freshwater production. The complex coupling of nonlinear channel flow along with mass transfer of water and salts across the membrane makes it challenging to determine an optimal spacer design. Conventional methods employ a design loop based on manual design and re-evaluation, which is time-consuming and typically will only find an improved, but not an optimal design. This paper presents an alternative approach that employs the gradient-based adjoint method to explore an optimal design in complex shapes with many design parameters. The novel design methodology provides an effective tool for designers to pursue advanced spacer designs. To elucidate the design method with sensitivities, the spacer in cavity, submerged and zigzag configurations are first analysed with the computed gradients of pressure loss and flow permeation with respect to the displacement of the grid nodes on spacer surface. Then, an optimisation case of cylinder spacers in the zigzag configuration finds an optimal spacer that has 24% pressure loss reduction and only 0.43% permeation drop as compared to a standard cylinder spacer, indicating the new method's potential in finding high-performance spacer designs.