Automatic modulation classification has significant commercial and military applications and is one of the most challenging problems of the cognitive radio. To demodulate a captured signal, it is necessary to know the parameters of the signals and the preliminary information about the type of modulation. However, in most cases, the signal has no prior knowledge of the modulation type. Even more, there may not be a modulation pool available for possible modulation types. Cyclostationary-based features can be employed to blindly identify the unknown modulation type.