Early diagnosis of melanoma can substantially increase patient survival rate. Currently, dermoscopy is the dominant approach for clinical detection, but this method requires interaction with a trained clinical professional resulting in a financial burden which is a major limiting factor for many patients, especially those in remote and rural locations. It has been proposed that deep convolutional neural networks (CNNs) could allow an automated approaches for diagnosis of melanoma. However, there has been limited work regarding the use of CNNs to diagnose melanoma due to a limited amount of labelled training data available, a major limiting factor for the implementation of CNNs. This study utilises data augmentation techniques to improve CNN performance for diagnosis of melanoma, resulting a 12.4\% increase in validation accuracy despite the collection of no additional training data.