This letter considers a multi-channel access problem in a cognitive radio network with M primary users and a multi-channel secondary user (SU). SU selects K channels for transmission without knowing the channel states or statistics of channel evolution processes. If SU senses that a selected channel is available, it sends data over that channel at that time. Otherwise, SU does not use it in that slot. Under average throughput criteria, myopic policy is proposed and upper bounds of its throughput performance are derived for this problem. It is shown that it achieves optimality over block fading channels for certain channel evolution processes although it is generally suboptimal.