In order to be able to draw relevant conclusions, several experiments have been carried out using two challenging and realistic face image databases called Labeled Faces in the Wild (LFW) [14] and the so-called Gallagher’s database [10], where the images were taken in unconstrained conditions. Our Local-DNN framework outperforms other deep learning methods evaluated in this work, such as standard DNNs and DCNNs, and also obtains state-of-the-art results on these databases.