We formulate a mathematical version of the conventional model of maintenance of silencing in S. cerevisiae and analyze the conditions for bistability as well as for formation of stationary boundaries. Although the model is perhaps too simple, the structure of the bifurcation diagram, describing parameter regions with different kinds of qualitative behavior, is likely to be more robust. We can place some of the known mutants in different regions of this diagram. One interesting finding of this study is that, under some conditions, the lowering of acetylation rates might have non-obvious consequences for silencing. Possible improvements of the model as well as experimental tests are discussed at the end.