The neural substrate of working memory is an important issue in neuroscience. An important candidate for this at the moment is delay activity. A model of delay activity was given by Amit & Brunel (1997). Recently, Knight and others introduced a novel simulation technique that describes the evolution of a state probability density function of a population in terms of its external input, from which its firing rate can be determined. We apply this technique to a model which is motivated by the original model of Amit and Brunel and try to describe dynamical aspects of working memory