Automatic classification of the modulation type of an unknown signal has an important role in communication systems, especially in civilian purposes and in communication intelligence and military applications, to extract useful information from the signal. It has emerged in the research of communication systems, particularly in the study of software radio, due to the advances in reconfigurable signal processing systems. It is necessary for a multipurpose receiver to automatic choose among modulation types, in the presence of noise, in order to extract the features from the signal without distorting the signal in a manner that would affect the information that may be of interest and this is indeed a difficult process. A wide variety of techniques and approaches, for determining the modulation type applied on the signal, have been cited in the literature and this research will address the problem of automatic modulation recognition of analogue and digital communication signal using three methods i.e. Statistical Signal Characterization (SSC), Entropy of the Continuous Wavelet Transform (CWT), and Linear Predictive Coding (LPC) of the CWT. The first technique is a new method and it uses the SSC idea to classify the different modulations of the signal. It extracts four different parameters from a certain modulated signal and compares these parameters to those of other different modulated signal. This provides a basis for the classification system proposed in this work. Artificial Neural Network (ANN) was used, with the four SSC parameters, to classify the modulation type of the input signal. This robust classification system can identify the modulation types with a low Signal to Noise Ratio (SNR) of 6 dB with an efficiency of 98% and 97% for analogue and digital modulations respectively and at 3 dB with an efficiency of 83% and 86.6%. The other two methods mentioned above for modulation classification have been reported in the literature and this research is reporting an extension work on them to identify more modulation types. Continuous Wavelet Transformation (CWT) was employed in these two methods with different wavelets to retain the time variable information in the signal, which is very important in the classification. The two methods were combined to form a hybrid system or multiple network classifier that was compared with Azzoz and Nandi [4] hybrid classifier system. It can identify the modulation type with SNR down to 6 dB with an efficiency of 90% and 95% for analogue and digital modulations respectively. A modulation level classifier is also proposed in this work to classify level 2 from level 4 of 4 digital modulations. The input to the designed systems is either analogue or digital signals, or combination of both. There are five analogue and three digital modulations were used for identification by the three methods. Finally, the methods were compared together regarding to some different parameters. The SSC technique is simple and shows best results at lower SNR compared to the other two methods.