Damage identification is an important component for accurate lifetime predictions of any structure. In the case of a composite structure, however, damage can occur at several material scales: it can vary from micro damage, like fiber debonding or micro-cracking, to global damage such as buckling or delamination. These different material scales make damage identification difficult with a single type of sensing device. A single embedded optical fiber, causing little perturbation to the surrounding host structure, can multiplex hundreds of sensors, and furthermore, sensors measuring at different length scales. For example, short Bragg gratings can measure strain at given locations; long Bragg gratings can measure strain gradients; interferometric techniques can measure integrated strain along a given fiber length. The use of multi-scale measurements has been shown by the authors to improve the precision of damage identification. Still the treatment and fusion of these data is a non-trivial problem. This work presents a back propagation Neural Network algorithm used to fuse simulated multi-scale sensor data in order to identify damage. An analytical model of an isotropic plate subjected to a known load and specific forms of damage is used to train the network. The input data are: localized strain, localized strain gradient, and integrated strain measurement along a regularly spaced sensor network. This method is tested against a randomly generated set of damages. The combined use of multi-scale measurements and Neural Network analysis shows a great potential in damage identification for composite structures.