Two-stage design is very useful in clinical trials for evaluating the validity of a specific treatment regimen. When the second stage is allowed to continue, the method used to estimate the response rate based on the results of both stages is critical for the subsequent design. The often-used sample proportion has an evident upward bias. However, the maximum likelihood estimator or the moment estimator tends to underestimate the response rate. A mean-square error weighted estimator is considered here; its performance is thoroughly investigated via Simon's optimal and minimax designs and Shuster's design. Compared with the sample proportion, the proposed method has a smaller bias, and compared with the maximum likelihood estimator, the proposed method has a smaller mean-square error. Copyright © 2010 John Wiley & Sons, Ltd.