A dynamic environment, viz. , that of a computer network, poses a problem which conventional clustering techniques cannot handle. A weighted, dynamic graph is used to model the problem environment. The clustering problem can be formulated so as to achieve different objectives, and many of these possibilities are discussed. The problem of clustering is shown to be NP-complete in most of its formulations. Top-down and bottom-up approaches to clustering are proposed. The latter approach is developed in detail and a taxonomy of bottom-up algorithms is given. The basic paradigm for the bottom-up approach is given and its time complexity is analyzed. Algorithms are described that are especially suitable to a distributed environment. The technique is applied to the clustering problem in the dynamic packet radio environment. Extensive simulations have been carried out. Results are reported and various heuristics are compared along the dimensions proposed in the taxonomy. A short discussion on distributed clustering is given.