A novel method for detection of ventricular late potentials (VLP) using artificial neural network (ANN) from wavelet preprocessed vector magnitude waves (VMW) is proposed. The VMW is firstly processed with a continuous wavelet transform (CWT). Then eight features are extracted from time frequency energy distribution of VMW, and are inputted into ANN for VLP detection. The ANN is trained with 40 clinical samples and tested using another 38 clinical samples, respectively. The results show that the specifically designed ANN can detect VLP with the high rate of correct classification (93.33%), and can enhance the sensitivity and specificity of VLP detection as compared with conventional time domain method. A novel method for detection of ventricular late potentials (VLP) using artificial neural network (ANN) from wavelet preprocessed vector magnitude waves (VMW) is proposed. The VMW is once processed with a continuous wavelet transform (CWT). Then eight features are extracted The results show that the specifically designed ANN can detect VLPs with VLP with the high rate of correct classification (93.33%), and can enhance the sensitivity and specificity of VLP detection as compared with conventional time domain method.