Energy disaggregation estimates appliance-by-appliance electricity consumption from a single meter that measures the whole home's electricity demand. Compared with intrusive load monitoring, NILM (Non-intrusive load monitoring) is low cost, easy to deploy, and flexible. In this paper, we propose a new method, coined IMG-NILM, that utilises convolutional neural networks (CNN) to disaggregate electricity data represented as images. CNN is proven to be efficient with images, hence, instead of the traditional representation of electricity data as time series, data is transformed into heatmaps with higher electricity readings portrayed as 'hotter' colours. The image representation is then used in CNN to detect the signature of an appliance from aggregated data. IMG-NILM is flexible and shows consistent performance in disaggregating various types of appliances; including single and multiple states. It attains a test accuracy of up to 93% on the UK dale dataset within a single house, where a substantial number of appliances are present. In more challenging settings where electricity data is collected from different houses, IMG-NILM attains also a very good average accuracy of 85%.