The literature nonstoichiometry data for bulk defects in LaxSr1-xMnO3 +/-delta (LSM) has been analyzed using a rigorous inverse-problem methodology. A solver-optimizer combination of Newton's method and the particle swarm optimizer (PSO) was developed, along with a rigorous method for determining local identifiability in the comparison between defect models and experimental data. It was found that previous models, which do not account for any excess free energy for defects, cannot adequately replicate nonstoichiometry data at low temperatures. A revised model is proposed that includes excess free energy due to strain field interaction between cation defects. The revised model was shown to be consistently distinguishable from the earlier model (a better fit) even when using a model structure containing fewer parameters. It was also identifiable for 10 and 20% Sr-doping. Data for activation energies and vibrational entropy for the various defect reactions was compiled. (C) 2008 Elsevier B.V. All fights reserved.