In this paper, we address the problem of identifying the modulation level of the received signal under an unknown
frequency selective channel. The modulation level classification is performed using reduced-complexity Kuiper (rcK) test which
utilizes the distribution of signal features such as magnitude of the received samples or phase difference in consecutive
received samples. However, in frequency selective channels, these features are severely distorted resulting in a poor classification performance. We propose to use constant modulus algorithm (CMA) to mitigate the impact of the frequency selective channel on the signal feature. Simulation and analytical results show that the proposed CMA-rcK technique outperforms state-of-the-art cumulant-based technique as well as blind equalizer-based technique that uses Alphabet Matched Algorithm.