Gait training is a major part of neurological rehabilitation. Robotic gait training systems provide paraplegic patients with consistent, labor-saving, and adjustable physical therapy over traditional manual trainings. However the high cost and social-technical concerns on safe operation currently limit their availability to only a few large rehabilitation institutions. This paper describes the synthesis of a linkage mechanism for gait pattern generation in a sagittal plane. The synthesis of the mechanism starts with the definition of a closed ankle trajectory obtained from normative gait data. The synthesis process we developed includes (1) construction of the desired ankle trajectory, (2) formulation of an objective function to be used for linkage optimization, (3) development of a procedure for transforming an initial guess to a starting set of design variables for optimization, and (4) development of a point-matching process needed for implementation. A set of stature-referenced parameters was successfully produced for a crank-rocker mechanism to generate the desired gait path. A simple linkage mechanism can be used as the pattern generator in a gait training system, and the presented process has been used to synthesize a linkage for a specific gait pattern.