Optimal tolerance design aims at assigning tolerances such that the functionality requirements are achieved with minimum cost. Classical tolerance design procedures are based on the assumption of independence of variables. This assumption might not be realistic, leading to the assign of non-optimal tolerances. This paper proposes an innovative methodology to allocate optimal statistical tolerances to dependent variables, where the dependence structure is estimated from the manufacturing process. The methodology is based on the assumption that the multivariate process variability is consequence of a set of independent factors. Hence, the tolerance assignment should be based on the statistical properties of these factors. The tolerance design takes the independence of the factors as a restriction, and tolerances are optimally assigned according to the variability of them. © 2008 Elsevier Ltd. All rights reserved.