In this paper a system for recognizing pulse compression radar waveforms is introduced. The waveforms considered in this study are the linear frequency modulation (LFM), Costas codes, binary phase codes, and the Frank, P1, P2, P3, and P4 codes. A feature vector based on instantaneous signal properties, second- and higher-order statistics, and time-frequency distributions is computed from the received signals. Cyclic correlations are used in symbol rate estimation. Information theoretic measure is used to remove redundant components from the feature vector. The discrimination capability of the features is evaluated using an ensemble averaging early-stop committee of multilayer perceptrons. Bayesian MLP classifier is considered as well. In simulation the classifier attains over 97 % overall correct classification rate in signal-to-noise ratio (SNR) of 6 dB