This paper introduces the challenge of cross spectral periocular matching. The proposed algorithm utilizes neural network for learning the variabilities caused by two different spectrums. Two neural networks are first trained on each spectrum individually and then combined such that, by using the cross spectral training data, they jointly learn the cross spectral variability. To evaluate the performance, a cross spectral periocular database is prepared that contains images pertaining to visible night vision and near infrared spectrums. The proposed combined neural network architecture, on the cross spectral database, shows improved performance compared to existing feature descriptors and cross domain algorithms. © 2014 IEEE.