A set membership estimator (SME) based on limited number of measurements is proposed for estimating metabolite concentrations using dynamic flux balance models (DFBMs). To deal with multiplicity of solutions of the DFBM, a weighted primal dual method is used to find solutions that best fit the data. Multiparametric nonlinear programming is applied to propagate uncertainty in initial concentrations along a batch/fed-batch operation. The proposed method has been applied to E. coli batch and fed-batch fermentation without noise.