In the semantic analysis, I identified some 150 editing incongruencies in the sample. These grouped under five classes of inaccuracy, and we see how specific editing rules tend to result in certain kinds of inaccuracy. We can move from this description to prescriptive guidelines on what changes editors should and should not make. Analysis of which stories were in- accurately edited shows that certain categories of news suffer more than others. We can quantify the seriousness of inaccuracies, and relate these to various extra-linguistic factors. Do certain individuals or stations edit less accurately? Are certain kinds of news more liable to inaccurate editing? I present evidence on these .questions in the last section of this paper.