In survey research, numerous studies deal with best practices concerning the optimal design of rating scale items commonly arranged in grid formats. Taking account of mostly indirect indicators of data quality including response times, nondifferentiation and other forms of response styles various advice exist regarding an appropriate construction of rating scales (e.g., in terms of number and labeling of scale points, offering a midpoint or a 'don't know' option). Field-experimental results concerning these design features, however, are often mixed and the cognitive processes underlying respondents' answers to rating scales remain unclear. Eye tracking may help to gain a better understanding of the cognitive processing of rating scales and its interaction with different design features. In a lab experiment, university students from various disciplinary backgrounds (n=150) are randomly assigned to different rating scale designs presented as part of a Web questionnaire. By analyzing the respondents' eye movements, different scale formats (e.g., 5-point scale vs. 9-point scale, with or without midpoint, nonsubstantial option provided or not) are compared concerning the amount of attention and cognitive effort respondents spend on processing various components of the rating scale. Eye tracking data are assessed in order to ascertain whether differential design properties of rating scales evoke differences in the degree of visual attention and in the scanpath of respondents which ultimately can be held responsible for differences in survey responses provided by respondents.