Recent progress has been made in estimating prognosis in primary biliary cirrhosis using Cox models. These models have also demonstrated the therapeutic value of liver transplantation by comparing the observed survival for a group after transplantation with the expected survival without transplantation calculated from the Cox prognostic model. However, good risk patients and those not transplanted principally for hepatocellular failure may not have a survival advantage for many years. Cox models have several limitations: the selection criteria for the patient populations used to derive the models, the selection of the time at which the patients are evaluated, the poor prognostic accuracy for individual patients rather than patient groups and lastly the fact that they use variables derived at only one time point-time independent Cox models. Thus new statistical tools must be used to improve prediction of survival in individual patients with PBC in order to optimize timing of liver transplantation. In addition a more precise definition of the natural history of both symptomatic and asymptomatic forms of this disease is needed to evaluate the efficacy of therapeutic agents in randomized clinical trials. However, although use and timing of therapeutic intervention, including liver transplantation, still requires good clinical experience and judgement, statistical modelling does give some objective measurement of prognosis, which is useful for the clinician treating patients with PBC. At the same time that new treatments are being evaluated, there is an obvious need to improve prognostic tools for application to individual patients with PBC. This may be achieved by using serial data in a different form of modelling-time dependent Cox models.