A study was conducted to investigate the use of time-frequency methods to predict the outcome of defibrillation for patients presenting with ventricular fibrillation. Both wavelet transform and short time Fourier transform methods were applied to extract characteristic features from a data base of pre-shock signals. A Bayes classifier was developed for classifying between those outcomes where a return of spontaneous circulation (ROSC) was achieved and those where it was not (NOROSC). Probability distribution functions were estimated using multidimensional histogram and Gaussian kernel smoothing techniques. Cross validation was employed to improve the confidence of results. Three formats of feature sets including the original feature sets, normalised feature sets, and principal component analysis (PCA) feature sets were used in the classification. The optimal pre-shock length and temporal location were investigated. In related studies the a posterior probability function was employed to indicate the probability of successful shock (PROSC) and the effect of wavelet central frequency was also studied. The best classification performance for the original, normalised, and PCA feature sets were 58±2% specificity at 90±4% sensitivity, 59±3% specificity at 90±4% sensitivity, and 56±3% specificity at 92±4% sensitivity respectively. Overall it was found that the analysis employing time-frequency-based methods improved the performance of shock outcome prediction when compared to currently available alternative methods.