Although ordinal rating scales have received much research attention in survey methodology literature, the direction of the rating scales has not been as extensively studied as other design features. Research on scale direction effect has mainly focused on the influence on response distribution, while largely overlooking its impact on latent constructs. This study examines the scale direction effect on extreme and acquiescent response style latent class variables in an experiment embedded in a national probability sample. We found a higher level of acquiescent response style from scales starting with positive adjective words using a web survey. No significant effect of scale direction was detected on extreme response style or in a face-to-face survey (with show cards). This study also demonstrates that scale direction does not affect the substance latent class variables, once the response style latent class variables are included in the model. Implications of these findings and future research directions are discussed.