The proposal of the present work is that the brain selects stimuli for further processing based on their saliency, defined as the strength of their neural representation. A saliency-based model of attentional selection is proposed consisting of two hierarchically arranged processing stages. At the first stage basic features of stimuli are represented in distinct feature maps, which are subject to top-down modulation from working memory. Feature maps feed into the saliency map, which represents the second stage of processing, and create distinct saliency differences in this map. An important feature of the model is that saliency is affected by physical characteristics of stimuli such as brightness or colour as well as by top-down influences such as behavioural goals, and that at the level of the saliency map these distinct influences are indistinguishable. The model proposes that attention selects the most salient stimulus for further processing.