AbstractDomain knowledge is the soul of software systems. After decades of software development, domain knowledge has reached a certain degree of saturation. The recovery of domain knowledge from source code is beneficial to many software engineering activities, in particular, software evolution. In the real world, the ambiguous appearance of domain knowledge embedded in source code constitutes the biggest barrier to recovering reliable domain knowledge. In this paper, we introduce an innovative approach to recovering domain knowledge with enhanced reliability from source code. In particular, we divide domain knowledge into interconnected knowledge slices and match these knowledge slices against the source code. Each knowledge slice has its own authenticity evaluation function which takes the belief of the evidence it needs as input and the authenticity of the knowledge slice as output. Moreover, the knowledge slices are arranged to exchange beliefs with each other through interconnections, i.e. concepts, so that a better evaluation of the authenticity of these knowledge slices can be obtained. The decision on acknowledging recovered knowledge slices can therefore be made more easily. Our approach, rooted as it is in cognitive science and social psychology, is also widely applicable to other knowledge recovery tasks. Copyright © 2001 John Wiley & Sons, Ltd.