Numerous learning curve models have been offered in the literature and used in practice. This paper selects five learning curve models which differ with respect to the pattern of learning assumed to exist, and investigates the forecast accuracy of the varying models under varying circumstances. The broad objectives are to identify conditions which may affect model accuracy, documenting the manner in which forecast errors for each model depend on those conditions, and suggest which of the five models may be more or less accurate under a given set of conditions. Tests are conducted using annual cost and quantity data from a large sample of major aerospace weapon system programs. Learning curve, Cost progress model, Cost estimation, Forecasting accuracy.