Multihop packet radio networks require routing algorithms which are distributed in nature and which have the ability to timely detect changes in the network topology. These changes are mostly changes in connectivity caused by link or node failures and by the relative motion of the nodes. This paper describes and analyzes an adaptive decentralized routing algorithm for packet radio networks. The network connectivity, as perceived by each node, is translated into a graph representation of the network. The proposed routing mechanism then uses a breadth-first search algorithm along the inbound links of such a graph. Unlike most of the algorithms found in the open literature the one proposed here can be used in networks having both uni- and bi-directional radio links. Examples are shown to illustrate the methodology. >