Spasticity is a hypertonic muscle behavior commonly observed in patients with multiple sclerosis, cerebral palsy, stroke, etc. Clinical assessment for spasticity is done through passive stretch evaluations of various joints using qualitative clinical scales, such as the Modified Ashworth Scale (MAS). Due to the subjective nature of this evaluation method, diagnostic results can have poor reliability and inconsistency. A few research groups have developed electromechanical training simulators of upper arm spasticity with the intent of providing healthcare students practical training opportunities. This paper presents a novel, purely mechanical (nonpowered) training simulator as an alternative design approach. This passive design utilizes a hydraulic damper with selectable viscous effect to simulate the speed-dependent spastic muscle tone and a Scotch-Yoke linkage system to create the “catch-release” behavior of spasticity. An analytical fluid model was developed to systematically design the hydraulic damper. The error residuals between model prediction and experimental damping force were found within ±2.0 N and percent errors within ±10% across various testing speeds (i.e., 250, 500, 750, and 1000 mm/min). The performance of the fully assembled simulator was tested under slow (ω ≤ 60 deg/s), medium (60 deg/s < ω < 150 deg/s), and fast (ω ≥ 150 deg/s) stretch speeds, where ω is the joint angular speed. Preliminary bench-top results suggested the feasibility of replicating five distinct levels of spasticity behaviors (MAS levels 0–4), where resistive torque increased with higher stretch speed and peak resistive torque ranged from 1.3 to 6.7 N · m under the fast stretch speed.