Communication system recognition can be used in some civilian and military applications. The recognition of the system is done by inspecting the received signal properties like modulation type, carrier frequency, baud rate and so on. Therefore we need Automatic Modulation Recognition (AMR) in addition to carrier and baud rate estimation methods. In this paper we introduce a new AMR method based on time and spectral domain features of the received signal. A neural network is used as the classifier. A broad class of analog and digital modulations is considered. Baud rate and carrier frequency estimation is performed by existing methods referred to in this paper. Using this information the protocol used for signal transmission is detected.