At a time where customer preference is becoming an important product development criteria, measures which quantify subjectively perceived auditory sensations are becoming useful in developing meaningful sound quality criteria. One proposed measure which has not yet seen a great deal of application to automotive sounds is that which attempts to quantify the sensation of roughness. The applicability of such a proposed measure can be established through a series of subjective experiments. Typically, such experiments involve the presentation and evaluation of a group of sounds which vary in their degree of roughness. In order to generate test sounds, a system for the modeling and synthesis of pawertrain sound has been developed which isolates specific signal components which are known to affect the roughness of a sound. The system provides the ability to precisely adjust specific characteristics which are known to affect the sensation of roughness without significantly changing other signal characteristics. The models used by this system are derived from actual powertrains, and can simulate realistic steady state as well as transient behavior such as wide open throttle acceleration. The parametric engine model used in the system consists of a 20h order harmonic component, a stochastic ‘noise’ component, an order-tracking transfer function component, and an AM/FM modulation component. The design and capabilities of the system are reviewed, and preliminary results of a subjective study of in-line four cylinder engines using synthesized stimuli are presented.