This paper presents a frequentist comparison of the performance of confidence and credibility intervals for the difference of two proportions from two independent samples. The comparison is carried out considering three frequentist criteria. It was found that the intervals with the best performance, in terms of coverage probability, are Bayesians; in terms of expected length and variance of the length, the Newcombe interval shows the best performance. As a final remark, it was found that traditional intervals such as the Wald and adjusted Wald have a poor performance.