Spasticity is a human motor system disorder in which reflexive muscle activity becomes unregulated, causing unwanted contractions that can interfere with voluntary movement. We present a simulator that replicates spastic arm dynamics for clinical training of physical therapists and neurologists. Accurate clinical assessment of spasticity is critical in the determination of patient treatment, although physical evidence of spasticity is often confused with that of related neuromuscular disorders. By repeatably simulating different levels of spastic severity, we hope to improve clinician training for rating spasticity and consequently decrease the variability of ratings between raters and within raters. Our haptic device, designed to replicate the spastic elbow of a child, uses a brake actuator and high-resolution optical encoder. Two competing spasticity models from the literature are implemented. Preliminary experiments indicate that the decreased stretch reflex threshold model is more realistic than the increased stiffness model. The simulator improves on training with patients, since spastic severity can be readily adjusted under controlled and repeatable conditions.