The invention particularly relates to a digital modulation signal classification method based on union features, and the method comprises the following steps: (A) obtaining an intermediate frequency real signal after performing ADC sampling for an intercepted signal by a receiving end; (B) preprocessing the intermediate frequency real signal to obtain an intermediate frequency complex signal; (C)obtaining an instantaneous feature quantity, a higher-order statistics and a spectrum feature of the intermediate frequency complex signal, and identifying the signal according to the features, if theidentified signal is a 2ASK signal, a BPSK signal, a QPSK signal, a OQPSK signal and a 8PSK signal, ending the classification; if the identified signal is a MFSK or MQAM modulation main class, executing next step; (D) performing line class internal identification for the MFSK or QAM modulation main class, identifying out a modulation order M in the current modulation main class and then ending the classification. The method combines various features to perform modulation identification, classifies the signals of which features are approximate through multi-level spectrum features, and therebyhas high identification rate and wide range of application.