This study develops a set of preventive maintenance policies, based on the idea of critical machines, for use in group technology shops. These policies designated certain machines in the shop as critical. When a critical machine breaks down, it is repaired and preventive maintenance is applied to a block of machines (either the remaining machines in the cell, the other machines of the same type or all of the remaining machines in the shop). Simulation of a complex group technology shop was used to compare twelve definitions of criticality and three preventive maintenance conditions. The definition of criticality was not found to be a major cause of performance differences between shops, since in many cases, the same machines were critical by any definition. The application of preventive maintenance to all machines in the shop when a critical machine had failed led to the best shop performance, indicating that the reduction in machine interference caused by doing so was more important than the unique features of group technology.