Fixed threshold based conventional energy detection methods may not work well in low signal-to-noise (SNR) cognitive radio (CR) scenarios where the ability to detect a primary user or to avoid a false alarm has been found to be severely degraded. In this paper, we investigate the effect of low SNR on a static threshold based energy detector. We discuss the reasons for the degradation in performance of the CR at low SNR. We also propose a method to change the threshold depending on the SNR of the channel to make the response of the CR suited to the operational requirement. Simulation results show that adaptive setting of the threshold is effective in mitigating the deterioration of CR performance metrics at low SNR.