A classification architecture is presented in which the domain knowledge and belief measures are represented using the principles of possibility theory. The representations and support generation strategies are designed for a classification system whose objective is to identify radar types from signals received by passive sensors. A measure of belief is generated to indicate the support for each of the alternatives on the basis of the acquired evidence, using a possibility distribution over the frame of discernment. The immediate transformation of evidence to possibility distributions avoids the computational difficulties associated with utilizing the joint possibility distribution over the entire set of attributes that characterize the domain objects. The desired possibility distribution is obtained by combining the compatibility measures for each hypothesis independently.