Attitude measurement in web surveys mostly relies on requiring respondents to indicate their agreement or disagreement with each of several items under the same Likert-type scale in a grid format. As it is known that respondents do not only attend to the words that convey the questions but also to the visual language of a questionnaire (i.e., format and shape of response scales, verbal and numerical labels of scale points, spacing, positioning, and order of response options) it is essential to understand how this effects the response process. Although there is no conclusive evidence about the influence of the direction of extreme point labeling (e.g., Belson, 1966; Friedman et al., 1993; Salzberger & Koller, 2010; Weng & Cheng, 2000), applying the ―near means related‖ heuristic (Tourangeau et al., 2004; 2007) to horizontal Likert-type scales would suggest that the proximity between the item and the positive anchor of the scale in a agree-to-disagree format would lead to different results than a reversed scale (disagree-toagree format). This study aims to bring forward how the direction of Likert-type scales in grid formats influences the response behavior of different respondent groups. In two independent web surveys with online panel members and professionals respondents were assigned to one of four treatment groups. The direction of Likert-type scales as well as the use of numerical labeling of scale points was experimentally varied in a full-factorial 2 (agree-to-disagree vs. disagree-to-agree) x 2 (with numerical labels vs. without numerical labels) design. The influence of scale presentation was measured on different indicators of data quality (response latency, item omission, non-differentiation in grids, response sets).