Reliable assessment is essential for the management of spasticity, one of the most frequent complication of various neurological diseases. For the spasticity assessment, several clinical tools have been developed and widely used in clinics. The most popular one is modified Ashworth scale (MAS). It has a simple protocol, but is subjective and qualitative. To improve its reliability, quantitative measurement and consistent training would be needed. This study presents an elbow spasticity simulator which mimics spastic response of adult post stroke survivors. First, spastic responses (i.e. resistance and joint motion) from patients with a stroke were measured during conventional MAS assessment. Each grade of MAS was quantified by using three parameters representing three characteristics of the spasticity. Based on the parameters, haptic models of MAS were developed for implementing repeatable and consistent haptic training of novice clinicians. Two experienced clinicians participated in preliminary evaluation of the models.