Cost-effectiveness analysis (CEA) has a relatively short history in health and medical care, reflecting the fact that costs have only recently been considered an important factor in medical decisions. Not long ago, the only relevant question was whether a particular therapy worked (that is, whether it was effective); now, once a therapy has been shown to be effective, the next question is whether it provides sufficient value to be worthwhile (that is, whether it is cost-effective). It is this second question that CEA attempts to address, quantitatively and objectively, by measuring all the relevant financial costs and all the relevant health effects of a therapy, then weighing these results explicitly in a cost-effectiveness ratio. The pioneering studies that introduced CEA into the medical field applied mathematical models to data about the costs and outcomes of various therapies, data drawn from the literature; the models and data were combined to calculate the cost-effectiveness of the therapies.1-4 These early studies were limited by several factors.The synthesis of information from a wide array of data sources necessarily injected a degree of investigator judgment into the assessment of cost-effectiveness. Then, too, data on costs and data on outcomes were usually drawn from different sources, thus raising questions about the comparability and consistency of that data. Also, the reporting of data in the literature was often sketchy and incomplete, especially when compared with the very specific requirements of the economic model being applied; the investigator was therefore forced to exercise his or her judgment to fill in the gaps left by the literature.The models used to calculate costeffectiveness were often complicated and therefore difficult to describe completely in a journal article, which raised concerns that investigator bias might have crept into the evaluation process before the results were published. Cost-effectiveness models were thus regarded as less objective and less reliable than other sources of information.5 In the meantime, randomized clinical trials began to measure the costs of therapies in addition to determining standard clinical outcomes.6-8 Investigators could now evaluate both cost and effectiveness outcomes in the same population of patients, with the protection provided by randomization that treatment allocation bias had been controlled.As a result of the meticulous planning needed for randomized trials, all data relevant to the CEA could be collected prospectively and in sufficient detail to provide the basis for a complete analysis. The strengths of randomization, prospective data collection, and objective measurement of outcomes represented a major advance in CEA.These factors also seemed to provide an answer to critics concerned about the issue of investigator subjectivity that was associated with the use of economic models.With the advent of real data, it was surmised, perhaps the need for models in CEA would no longer exist. Including cost measurement in randomized trials has indeed brought about a major advance in the field of CEA in medicine.The data that have resulted, however, have limitations that suggest that mathematical models will continue to be important in medical cost-effectiveness studies. Some of these limitations relate specifically to the techniques of cost-effectiveness studies.They involve such methodological issues as the projecting of study results beyond the end of observed follow-up, the placing of confidence intervals on cost-effectiveness ratios, and the manner by which cost-effectiveness studies may fail to account sufficiently for the effects on the cost-effectiveness ratio of variations in practice patterns and patient responses to therapy. Other potential limitations of cost-effectiveness studies in randomized trials relate to both clinical and economic outcomes.They involve questions about the generalizability of trial results to broader patient populations, about the distinction between the efficacy of a therapy in ideal settings and its effectiveness when applied in routine clinical practice, and about accounting for the variable of the “moving target”of changing medical technology. Each of these limitations can be addressed by the use of costeffectiveness models based on the observations from randomized trials.The remainder of this article will discuss each of these issues in turn, then return to the larger question about the future of models for CEA.