We present a hypothetical solution to the binding problem - a subject of fundamental importance for studies of cognition and consciousness. The solution is based on the idea of tagging neuronal messages and on the mechanism of Stochastic Diffusion Search. Tags allow to organise information processing to bind separate features into coherent, stable mental representations. Binding itself is performed dynamically by Neural Stochastic Diffusion Search Network (NSDSN). The neural correlates of the percepts are, in our model, dynamic rather than static patterns of activity. These mental representations - a form of working memory - could be either further integrated into the conscious stream or unconsciously stored in the memory for further recall. In the paper we will present NSDSN and its properties in the context of visual information processing. We will also discuss neurobiological evidence that lead us to the formulation of our model. Alternative solutions to the binding problem, proposed in the literature, will be also briefly discussed.