The paper focuses on computer aided diagnosis of heart diseases. We extracted QRS complexes of ECG signals from different patients. These ECG signals were decomposed using discrete wavelet transform. Using six different features of ECG signals were able to discriminate and classify six types of heart beats. Three of the features were statistically calculated from decomposed sub band signal. Two more features are taken as AC power and instantaneous RR interval of original signal. The effects of two wavelet decomposition structures, the two-stage two-band and the two-stage full binary decomposition structures, in the recognition of ECG beat types are studied. Either ANN or PNN are found to be useful to classify the features. Results show that two stage two band decomposition is sufficient and we can get very good accuracy by using just 11 features. The reason seems to be that most of the energy is concentrated in the Lower sub bands of the signals. The Higher frequency components do not have any notable information about diseases observable with ECG beat signals. © 2005 - 2013 JATIT & LLS. All rights reserved.