Abstract
This paper proposes a revised strategy of adaptive testing to circumvent a major assumption in the original strategy of adaptive testing that all remaining software defects are equally detectable in the whole process of software testing and the contribution of each single remaining defect to software failure rate keeps constant. In the revised strategy only fixed memory of latest testing data is used to update estimates of software parameters. As a by-product, the computational complexity of on-line software parameter estimation is substantially reduced. A case study with the real Space program shows that the revised adaptive testing is effective and can significantly outperform random testing.