In this paper, we first introduce the concept of cognitive machine-to-machine (M2M) communications. In cognitive M2M communications, devices generate packets and the packets are transferred to a concentrator. The concentrators are considered to be secondary users in the cognitive radio network opportunistically accessing a channel allocated to primary users. Based on a random access channel (RACH) mechanism, a queuing model to analyze the performance of cognitive M2M communications is proposed. The queuing model uses an iterative approach to reduce the complexity, and it is able to obtain various performance measures (e.g., queue throughput and delay).