The objective of this article is to demonstrate optimal adaptive design as a methodology for improving the performance of phase II dose-response studies. Optimal adaptive design uses both information prior to the study and data accrued during the study to continuously update and refine the study design. Dose-response models include linear, log-linear, 4-parameter sigmoidal E(max), and exponential models. Where the response has both a placebo effect and plateau at higher doses, only the 4-parameter sigmoidal E(max) model behaves acceptably and hence is used to illustrate the methodology. Across 13 hypothetical dose-response scenarios considered, it was shown that the capability of the adaptive designs to "learn" the true dose response resulted in performances up to 180% more efficient than the best fixed optimal designs. This work exposes the common misconception that adaptive designs are somehow "risky." As shown in this simple simulation example, the converse is true. Adaptive designs perform extremely well both when prior information is accurate and inaccurate. This leads to improved dose-response models and dose selection in phase III. This benefits sponsors, regulators, and subjects alike by reducing sample size, increasing information, and providing better dose guidance.