In this study, yogurt was fortified with iron and zinc in equal concentrations, 20, 40 and 60 mg per 1 kg of milk and the samples were tested in 1, 7 and 14 days of storage in terms of different properties of product including physicochemical properties (acidity, pH, syneresis, water holding capacity, viscosity), textural properties (hardness, elasticity, firmness, cohesiveness, adhesiveness) and sensory properties (texture, flavor, color, odor and acceptance). For predicting the changes in qualitative indices, neural network tool in MATLAB 2013Ra was used. In different networks, the Feed-Forward-Back-Propagation networks by 2-2-3-14 and 2-4-14 topology, with 0.997 and 0.991 correlation coefficients and 0.4090 and 0.1040 mean square errors, including hyperbolic tangent sigmoid transfer function, Levenberg-Marquardt learning algorithm and 1000 epoch was determined as the best neural models for yoghurt fortified with Fe and Zn, respectively. Optimal models were also investigated and the results of these models with high correlation coefficients (more than 0.98) and very low standard deviation were able to predict trends.