Learning curves and progress functions are well established management tools used to predict productivity in the start-up of new product lines, and to describe the performance of individual employees. As the authors, amongst others, have shown, these tools have been successfully applied to a wide range of tasks in highly varied industries. It is particularly useful, especially for inter-firm comparisons to compress the learning curves and progress functions into simple mathematical models in which the parameters may be determined by least squares error curve fitting or other convenient techniques. When on-line prediction is required, a digital computer algorithm is used to estimate the model parameters. It is essential that the parameter estimation technique used is robust in the presence of large amounts of scatter in the raw data. Particular problems arises in task in which the human operator is subject to job rotation, job re-design, or has to perform a variety of similar, but not identical tasks during batch production. Previous experience can therefore be regarded as prior practice in skill acquisition and it is important for management to determine the appropriate starting point on the new learning curve. Little has so far been published on this important topic. This paper will review the present state-of-the-art and show that in certain circumstances prior practice may he accounted for by relatively simple modification to learning curve. The results obtained are of particular relevance to production management in setting performance standards, production scheduling, labour cost evaluation and delivery date forecasting.