Efficiency in lean production systems is largely determined by the speed of recovery from process interruptions. Without buffer stock, and with a balanced line, the production rate will be constant except when a problem is detected, at which point the whole line will stop. This motivates exploration of problem solving as a key design feature of lean systems. One response to the need for improving operator problem solving skills has been the use of job rotation schemes. The increasing efficiency of operators faced with repeating tasks is widely accepted and has been characterized by the well-known learning curve. This learning curve is modified to describe problem solution times, incorporating forgetting effects and treating both general skill and specific skills related to a particular problem. The resulting model is explored in a simulation of a serial flow shop subject to a range of interruptions. The efficiency of the flow shop is characterized by its run-ratio and the effects of eight key variables on the run-ratio are tested through simulation. The results show that the run-ratio generally increases as operators learn more rapidly and forget more slowly, and decreases as the number of problem types increase. The effect of job rotation schemes, however implemented, is always to reduce the run-ratio. It is shown that this occurs because the effect of job rotation is to avoid any possible advantage in specialization related to problems which do not occur uniformly at all workstations.