We propose a method of modulation format identification based on compressed sensing using a high-order cyclic cumulant combined with a binary tree classifier. Through computing the fourth-order cyclic cumulant of the pretreated band signal, which is obtained by compressed sensing with the sampling rate much less than the Nyquist sampling value, the feature vector for classification is extracted. Simulations are carried out in the optical coherent fiber communication system with different modulation formats of multiple phase-shift keying and multiple quadrature amplitude modulation. The results indicate that this method can identify these modulation formats correctly and efficiently. Meanwhile, the proposed method is insensitive to laser phase noise and signal noise.