The paper introduces a knowledge-based statistical approach to tolerance allocation, where a systematic analysis for estimating process capability levels at the design stage is used in conjunction with statistical methods for the optimization of tolerances in assembly stacks. The analysis employs detailed knowledge on the tolerance capability of a wide range of manufacturing processes, including the effects of material and design geometry, enabling realistic tolerance distributions to be modeled for key characteristics on designs. The optimization of tolerance stacks is based around achieving the lowest assembly standard deviation for the largest possible and most appropriate component tolerances. The method takes into account failure severity through linkage with Failure Mode & Effects Analysis (FMEA) for the setting of realistic capability targets. The application of the method is fully illustrated using a case study from the automotive industry. The use of interactive software for supporting the tolerance allocation process is discussed and a number of areas are identified where further work is being directed.