Mobile-learning (m-learning), or mobile-assisted language learning (MALL), has been the object of a great deal of research over the last twenty years. However, empirical work in this area has largely failed to produce generalizable conclusions due to variation in methodology, target feature, and task-type (Burston, 2014, 2015). As schools in Japan begin to join the growing number of classrooms worldwide using mobile-based assignments, this study examined how Japanese EFL students' writing task production differed depending on writing medium (i.e., handwritten on paper vs. tapped on a smartphone). Writing samples were collected from N = 1,449 participants, divided into smartphone- or paper-based groups, across a spectrum of English proficiencies. Handwritten submissions were found to be significantly longer than those composed on a smartphone (p < 0.001, d = 0.54), with differences being more pronounced for learners of higher proficiency than lower ones. Significance and effect sizes steadily dropped from p < 0.001, d = 0.66 for advanced learners to p = 0.168, d = 0.38 for beginners. These results indicate that care must be taken in designing m-learning activities, and that students must be given adequate training in smartphone-input skills (i.e., tapping) and time to acclimate before using such tasks for high-stakes assessments.