Semantic priming involves a combination of automatic processes like spreading activation (SA) and controlled processes like expectancy and semantic matching. An alternative account for automatic priming has been suggested using attractor neural networks. Such networks offer a more biologically plausible model of real neuronal dynamics but fall short in explaining several important effects such as mediated and asymmetrical priming, as well as controlled effects. We describe a new attractor network which incorporates synaptic adaptation mechanisms and performs latching dynamics. We show that this model can implement spreading activation in a statistical manner and therefore exhibit all priming effects previously attributed to automatic priming. In addition, we show how controlled processes are implemented in the same network, explaining many other semantic priming results.