Neuronal decoding is the process of using mathematical and computational techniques to interpret and extract meaningful information from measured brain activity, mapping brain responses back to the stimuli, behavioral and/or cognitive events. We provide an example of how deep neuronal networks can be trained to decode behavioral information latent in the spiking activity of neural populations. For this, we analyze neuronal activity in the primary olfactory cortex of a mouse trained in a behavioral task where associations of spatial locations and particular odors are linked to rewards. Using artificial dense feed-forward and recurrent neuronal networks, we could successfully decode the spatial location of the mouse and its running speed, trial by trial. Moreover, despite not being traditionally involved in spatial cognition, olfactory cortex decoding accuracy reached similar performance to decoding in the hippocampus, a well-known hub for spatial information. These results indicate that complex and multiplexed information is already present at early sensory processing stages, supporting a distributed and decentralized view of brain organization.