Cost overrun in construction projects is an important problem that can have serious financial consequences. It also reduces the effectiveness of planning and decision making not only for the contractor but for all parties involved. The problem keeps occurring despite many attempts made over the years to improve cost estimation and forecasting methods. This paper focuses on analyzing cost overruns in Turkey and comparing the cost estimates produced using the traditional estimating method against cost forecasts produced using the reference class forecasting (RCF) method. RCF is used to predict the final cost of a project predicated on actual cost outcomes of accomplished similar projects in the same reference class. RCF has been so far used to address estimation bias in early project development before design is completed and before all risks are accounted for to help a client to produce a more realistic forecast of the final project cost. In this paper, however, it is used to produce a forecast of the final project cost based on the contract sum. The work described in the paper involves collection of contract sum and final project cost data from 420 completed building projects. In order to test the ability of RCF to produce realistic forecasts of projects' final costs, 75% of the data are used to produce the cost overruns distribution and to determine the required optimism bias uplift values for the building projects, while the other 25% of the data are used for testing the accuracy of the forecasts produced for the final costs of projects in the established reference class. The results have indicated that the required uplift values are in the range of 4 to 44%. For instance, if the acceptable chance of cost overrun is 50%, 65% of the test projects would have required only 4% uplift of the contract sum to produce accurate forecasts of the final costs. If, on the other hand, the acceptable chance of cost overrun is only 10%, 5% of the test projects needed to use a 44% uplift to enable more accurate final project cost forecasts for the contractor. Additionally, to check the accuracy of the test data results obtained through applicable optimism bias uplift values, three standard error measuresroot-mean-square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE)are used.