An optimized measurement recipe is determined by reducing the set of measurement technologies and ranges of machine parameters required to achieve a satisfactory measurement result. The reduction in the set of measurement technologies and ranges of machine parameters is based on available process variation information and spectral sensitivity information associated with an initial measurement model. The process variation information and spectral sensitivity information are used to determine a second measurement model having fewer floating parameters and less correlation among parameters. Subsequent measurement analysis is performed using the second, constrained model and a set of measurement data corresponding to a reduced set of measurement technologies and ranges of machine parameters. The results of the subsequent measurement analysis are compared with reference measurement results to determine if a difference between the estimated parameter values and the parameter values derived from the reference measurement is within a predetermined threshold.