Cognitive radio has been a research hotspot because of its promise to improve the utilization of the assigned but unused radio spectrum. The complexity of current networks has been a main drawback of network development, and so will the problem of cognitive radio networks. To solve the complexity problem, autonomic computing has been proposed to enable the network self-management. This paper puts emphasis on the autonomic cognitive radio nodes architecture for the purpose of establishing a cognitive radio network with autonomic computing property from bottom up. We model the cognitive cycle using autonomic computing principles and introduce an autonomic cognitive radio conceptual model. And then, an autonomic cognitive radio nodes architecture (ACRA) based on the proposed conceptual model and a realization method are proposed. ACRA can realize the cognitive radio function with autonomic computing property, which makes the collaboration and management in a network formed by this kind of node easier.