The paper discusses the lessons learned from several years of investigation into the performance of a large air traffic control data processing system. To achieve the twin requirements implicit in the title, the time between faults must be made as long as possible, whilst the time required to clear a fault must be reduced to the minimum. The practice of building up some degree of system redundancy with stand-by equipment is also one that can be employed to ensure that the occurrence of a fault does not cause an appreciable loss of system facilities.The experimental air traffic control system, on which the original findings were based, illustrates the feasibility of designing a high standard of reliability into data processing equipment. Extensive data on the performance of the system have since been analysed and used to improve further the reliability factor.