Most of American adults have a smartphone and an increasing number of them do not have any other form of high-speed Internet access at home. Asuccessful adaptation of web surveys requires not only readjusting the format of the questions to a smaller display that allows for direct interaction (i.e., touchscreens) but also accounting to the fact that surveys on smartphones are more likely to be taken in contexts with more environmental stimuli, and therefore with greater chances of distractions and interruptions. In this paper, we study the attention span of survey respondents on smartphones by comparing their responses and behaviors to a matched sample of computer survey respondents. In a collection of experiments, we manipulate both the announced and the actual length of the interview, and measure the reliability of their responses using a large battery of checks. Additionally, we measure a number of survey metrics, including breakoffs, taking temporary breaks, rotating devices, and switching devices. The contribution of our paper is two fold.First, we provide practical guidelines for the design of web surveys that are possibly taken on a multiplicity of devices. Second, we make a methodological contribution by presenting it in an setting designed to recover the optimal responses.