The roundworm Caenorhabditis elegans is an important model system for understanding the genetics and physiology of touch. Classical assays for C. elegans touch, which involve manually touching the animal with a probe and observing its response, are limited by their low throughput and qualitative nature. We developed a microfluidic device in which several dozen animals are subject to spatially localized mechanical stimuli with variable amplitude. The device contains 64 sinusoidal channels through which worms crawl, and hydraulic valves that deliver touch stimuli to the worms. We used this assay to characterize the behavioral responses to gentle touch stimuli and the less well studied harsh (nociceptive) touch stimuli. First, we measured the relative response thresholds of gentle and harsh touch. Next, we quantified differences in the receptive fields between wild type worms and a mutant with non-functioning posterior touch receptor neurons. We showed that under gentle touch the receptive field of the anterior touch receptor neurons extends into the posterior half of the body. Finally, we found that the behavioral response to gentle touch does not depend on the locomotion of the animal immediately prior to the stimulus, but does depend on the location of the previous touch. Responses to harsh touch, on the other hand, did not depend on either previous velocity or stimulus location. Differences in gentle and harsh touch response characteristics may reflect the different innervation of the respective mechanosensory cells. Our assay will facilitate studies of mechanosensation, sensory adaptation, and nociception.