In this paper, we are interested in cognitive radio networks (CRNs) whose operation does not rely on channel sensing. A spectrum server is responsible for collecting spectrum availability and location information from primary radio networks (PRNs), and broadcasting this information to cognitive radios. By subscribing to this broadcast, a CR knows about the spectrum opportunities without sensing channels. Spectrum opportunity under this paradigm presents a multi-level structure that generalizes the well-known channel-sensing-based binary structure. This multilevel structure reflects a microscopic spectrum opportunity for CRs, and can be exploited to increase the CRN throughput. Under this structure, we study efficient spectrum access in a multi-CR environment, with the objective of maximizing the network-wide utilization of spectrum opportunity. The difficulty of our problem comes from the fact that different CRs may decide the same channel to be available, but at different levels. Therefore, channel access needs to be carefully coordinated. Both centralized and distributed solutions are provided, supporting different modes of operation. Numerical results verify the accuracy of our algorithms and the significant gain achieved by the multi-level framework.