In the unstable wireless channel in Wireless Sensor Networks (WSNs), the packet loss rate may vary from time to time. It may also be very high in certain networks in the different time slots. The main problem when the packets are dropped at a sensor node is that it is very difficult to distinguish between the normal packet drop and the malicious packet drop. So, we propose a Reputation System which is aware of the channel with adaptive detection threshold (CRS-A) to detect selective forwarding attacks in WSNs. The algorithm will process the behavior of sensor nodes during the forwarding attacks according to the deviation between monitored packet loss and the estimated normal loss. To optimize the detection accuracy of CRS-A, we find the optimal threshold for forwarding detection, which is adaptive to the time- varying channel condition and the estimated packet loss probabilities of compromised nodes.. The selective forwarding attacks are often hindered by the normal packet losses, complicating the attack detection. Therefore, it is difficult to detect the selective forwarding attacks and improve the overall network performance. Most of related work focus on monitoring the packet losses in each transmission link and separating the nodes with higher packet loss rates from the data forwarding path. The other solutions are not effective in detecting selective forwarding attacks as that of the proposed technique since the main difficulty of attack detection is to distinguish the malicious drop from normal packet loss. The normal packet loss rate in the transmission link that is used should be considered in the forwarding evaluation.