A multi-damage localization method based on signal subspace fitting and BP neural network model is proposed in this paper for multi-damage monitoring on composite structures. Firstly, the neural network model is trained for damage number counting using a large amount of guided waves array signals. Then the array manifold matrix is constructed based on the guided waves propagation model of multi-damage. On this base, a cost function is used to fit the equivalence relationship between array manifold matrix and signal subspace to realize multi-damage imaging and localization. At last, to verify the effectiveness of the proposed method, the experiment of multi-damage localization on a carbon fiber composite plate is carried out. The accuracy rate of damage identification reaches up to 91.6%, and the mean localization angle error and distance error are 1.6° and 15.3mm respectively, which show the effectiveness of the proposed method.