Two-stage grinding processes in mass-scale manufacturing unit are usually too complex to optimize, due to large number of interacting process variables, between and within the stages. Furthermore, statistical design of experiment techniques, such as factorial design, fractional factorial and response surface design by sequential experimentations, to determine the exact optimal process design for the overall interdependent two-stage system, are sometimes too difficult to implement, if not impossible. In this context, considering each stage in isolation and determining individual optimal conditions may not result in an optimal process design, when the entire two-stage system is considered. The aim of this study is to apply empirical modelling technique based on direct observations, for prediction of a two-stage grinding process behaviour having multiple response characteristics of continuous variables, and determine overall optimal process design to meet the specific customer requirements. In order to achieve the above goal, the study proposes an integrated approach using multivariate regression, desirability function, and metaheuristic search technique. Three different metaheuristic search techniques, viz. real-coded genetic algorithm, simulated annealing, and a modified Tabu search based on novel Mahalanobis multivariate distance approach to identify Tabu moves, are employed to determining near optimal path conditions for an industrial case study of two-stage CNC grinding (honing) optimization problem, having various process and variable constraints. Computational study results based on different metaheuristics, and applied on the same two-stage optimization problem, show that the modified Tabu search performs better and also offer opportunities to be extended for other multi-stage metal-cutting process optimization problems.