This article tests whether collecting pile-sort data online produces results similar to those obtained with face-to-face methods. We collected pile sorts from 227 university students in the cultural domain of emotions. To test for mode and design effects, we randomly assigned participants to face-to-face or internet modes and to either a 15- or 30-item sort. We found no evidence of mode effects in semantic structure, but the level of agreement among respondents varied by mode, depending on the number of items in the task. Agreement was higher among online respondents in the 15-item task but among face-to-face respondents in the 30-item one. Because the central purpose of pile sorting is to elicit semantic structure, we conclude that online data collection is a viable option for most researchers, but the mode effect in level of agreement implies that other design elements may affect comparability of face-to-face and web-based methods.