We introduce a theory of scan statistics on graphs 
and apply the ideas to the problem of anomaly detection 
in a time series of Enron email graphs.