We propose a method employing mathematical programming and global optimization techniques for solving inverse problems arising in biological regulatory network (BRN) reconstruction. This problem consists in estimating unknown parameters of a model that describe the structure and dynamics of a biological system from a set of experimental observations and can be naturally cast as an optimization problem: choose the parameter values minimizing a given distance between the observed and estimated values of some observable variables. This minimization is subject to constraints derived from the models. Two significant examples are presented on how to handle different kinds of dynamics: pattern formation occurring from diffusable and non-diffusable gene products in the drosophila melanogaster morphogenesis and reconstruction of the gene regulatory network of arabidospsis thaliana based on the identification of stable sub-networks during morphogenesis.