The negative side of the 9-point hedonic scale is not fully understood. When used to evaluate negative attributes (for example, bitterness), this hedonic scale may yield poor results. We evaluated hedonic ratings as affected by scale types [9-point-categorical (CAT), line (LIN) and labeled-affective-magnitude (LAM)] and polarity (uni- compared to bi-polar), and attributes (positive compared to negative). We compared sensitivity and confounding effects [contrast+panelist effects=CP] of positive- and negative-attribute ratings among 3 scales, and compared effects of uni- (negative-side only) compared to bi-polar scales on negative-attribute ratings. Grape juices (classified as A=Liked, B=Moderate, C=Disliked) arranged in 4 sets (AB-AC, BA-CA, AC-AB, or CA-BA; the left sample served first) were evaluated (N=60 consumers) for overall-liking using 3 scales. Low-sodium chicken broths (classified as M=Mild- and S=Strong-bitterness) arranged in 2 sets (MS or SM) were evaluated (N=216) for bitterness using 3 scales. Balanced/randomized presentation was practiced. With bi-polar scales: 1) consumers better differentiated negative-attribute ratings; 2) CP was higher for positive- than negative-attribute ratings [5.47 compared to 0.11, 12.41 compared to 0.09, and 82.66 compared to 0.23, respectively, for CAT, LIN, and LAM; 3) LAM was more affected by CP. With negative-attribute ratings, CP of LAM was higher for uni- than for bi-polar scales. CP was more pronounced for liked- than disliked-samples, resulting in higher score fluctuation. CAT was more affected by contrast effects whereas LIN and LAM were more affected by panelist effects. Polarity effects were obvious for the mild-bitterness sample, showing significantly different results between uni- compared to bi-polar scales [3.91 compared to 6.39, 4.28 compared to 6.49, and 41.05 compared to 63.24, respectively, for CAT, LIN, and LAM]; all ratings from bi-polar scales were not on the negative-side. For the strong-bitterness sample, uni- and bi-polar ratings were on the negative side, with LAM having more consistent pattern. This study revealed some drawbacks of hedonic scales induced by scale polarity/types and attributes.