Recent studies have demonstrated the power of neural networks for different fields of artificial intelligence. In most of fields, as machine translation or speech recognition, neural networks outperform previously used methods (Hidden Markov Models, Statistical Machine Translation, etc.). In this paper we show efficiency of LeNet convolution neural network for sign language recognition. We evaluate different approaches on the Spanish Sign Language dataset where we outperform state-of-the-art results where Hidden Markow Models were applied. As preprocessing step we apply several techniques to get the same size of input matrix containing gesture information.