Almost every physical interaction generates high frequency vibrations, especially if one of the objects is a rigid tool. Previous haptics research has hinted that the inclusion or exclusion of these signals plays a key role in the realism of haptically rendered surface textures, but this connection has not been formally investigated until now. This paper presents a human subject study that compares the performance of a variety of surface rendering algorithms for a master-slave teleoperation system; each controller provides the user with a different combination of position and acceleration feedback, and subjects compared the renderings with direct tool-mediated exploration of the real surface. We use analysis of variance to examine quantitative performance metrics and qualitative realism ratings across subjects. The results of this study show that algorithms that include high-frequency acceleration feedback in combination with position feedback achieve significantly higher realism ratings than traditional position feedback alone. Furthermore, we present a frequency-domain metric for quantifying a controller's acceleration feedback performance; given a constant surface stiffness, the median of this metric across subjects was found to have a significant positive correlation with median realism rating.