Graphs of numerical data allow the representation and communication of quantitative relationships in many important fields. We have developed an accessible graphics system that allows people with visual impairments to create and explore auditory and tactile graphs of multivariate datasets that are typically represented as 2D colormaps or 3D surface graphs. In this paper we describe an experiment conducted with both visually impaired and sighted (but blindfolded) users, to evaluate how effectively auditory and tactile graphs produced by the system convey the amplitudes and widths of 3D Gaussian surfaces. With respect to discriminating surface amplitudes, the users showed no significant differences in performance using either the auditory or tactile graphs. With respect to discriminating surface widths, performance was significantly better with the tactile graphs than with the auditory graphs. Both user groups performed similarly, showing no significant differences in error rates or discrimination abilities. Finally we found an effect of surface data range on performance, with higher error rates for graphs of higher or wider surfaces. The results of these studies provide insights into performance and usability that should allow developers to create more effective accessible graphics systems.