The present invention relates to systems and methods for determining accurate estimates of punitive damages in bifurcated cases. The method increases the accuracy of punitive damages predictions through the use of mock juries that have completed the liability phase of the mock trial and have found the defendant liable. Prior art methods of estimating punitive damages typically skip the liability phase of the trial and ask jurors to simply assume liability, an assumption that ignores the fact that the experience of going through the process of finding the defendant liable changes the nature and dynamic of the jury in a fundamental and unpredictable manner. The method provided takes the totality of that experience into account, including the perceptions, coalitions, and antagonisms that develop during liability deliberations. The method includes assembling a pool of jurors, presenting the liability case, and assessing the results to determine whether each of the jurors is plaintiff-oriented or defendant-oriented. Based on these data, a punitive damages research jury is assembled from the pool, and this research jury deliberates the liability case to a verdict. The damages case is then presented to a research jury which has found for the plaintiff, after which the research jury hears the punitive damages case and deliberates to reach a decision on the amount of punitive damages.