Machine Learning approaches such as Reinforcement Learning (RL) can be used to solve problems such as spectrum sensing and channel allocation in the cognitive radio domain. These approaches have been applied to other similiar domains such as in mobile telephone networks and have shown greater performance than the static channel allocation schemes used. The objective of this research is to use an RL technique known as Q-Learning to provide a possible solution for allocating channels in a wireless network containing independent cognitive nodes. Q-Learning is an attractive algorithm for such a problem because of the low computational demands per iteration. Many or the current proposed techniques suggest using a negotiation policy between two nodes to decide on which channel each may use, however a considerable problem with this is the overhead involved in the negotiation involved between the nodes. This paper suggests an approach where each node acts as an individual independant node, with virtually no collaboration with the other nodes. Results have shown that using such a technique gives fast convergence on an optimal solution when correct rates are chosen. It has also shown that the algorithm is very scalable, in that as the network grows, the state-action space does not grow sufficiently to cause major memory or computational demands