This exploratory paper examines one class of decision related to learning and operations improvement: how and where to develop new manufacturing technologies as part of a program of continuous improvement. We examine the reasons for both the selection and the relative effectiveness of two methods for manufacturing improvement: learning by doing (in-process learning) on the shop floor and learning by development and experimentation away from the shop floor (off-line learning). We then explore how the development strategy selected affects the subsequent decision-making process, and how strongly initial choices about improvement methods affect later ones. Empirical data are presented from two Korean shipbuilding companies. While the structural characteristics of a plant, such as its layout and equipment, do appear to influence the improvement method it selects, we find that other factors, such as corporate philosophy and plant history also play an important role in the initial selection of improvement method. However, these determining factors fade in importance sharply over time, as the plant develops its ability to improve using that particular method. The fact the plant has practised improvement by one method (and, therefore, has often become relatively better at that mode of learning), begins to dominate the selection of future improvement techniques. Although this process is quite rational at each planning step, it can result in improvement processes in which small changes in initial circumstances and managerial choices cause large changes in the overall improvement path. The results also suggest that when managers choose any improvement methodology, they are not only learning about their manufacturing systems - they are choosing how their operation will learn, and indeed may lock themselves into that mode of learning for a longer duration than that of the current set of improvement projects. ？ 1998 Elsevier Science B.V.