Iterative backward and forward consecutive mean excision (CME/FCME) algorithms are efficient in suppressing and detecting narrowband signals. Being computationally relatively simple, they can be applied for spectrum sensing in cognitive radios. In this paper, theoretical limits of detection for the CME algorithms are derived. These limits predict whether a signal is detectable or not. More specifically, resulting signal-to-noise ratio (SNR) values at which the signals can be detected are presented. In addition, rules for choosing the optimal CME algorithm are conveyed. The validity of the presented limits of detection is confirmed with computer simulations.