Reinforcement learning has shown that, when combined with deep learning techniques, is able to provide solutions to complex and dynamic problems. Air traffic control is considered to be an problem of such nature, which bears the question to mind; Can reinforcement learning be used to solve the problem of air traffic control. This work explores the applicability of reinforcement learning to the air traffic control problem by setting up a distributed system for training and experience collection. The problem is formulated as a decentralized system. Each aircraft is modeled as an agent that uses local observations while being limited to heading changes only. During learning, information about the actions of surrounding agents are added. It is shown that for low air traffic density scenarios the model is able to provide collision avoidance and approach the correct runway under realistic limitations. However, due to the lack of global coordination and limited modeling of spatial relation between states this method is unable to solve more complex and higher air traffic density situations.