Granger and Hyung (2004), Diebold and Inoue (2001) and Smith (2005) demonstrate how long memory and structural change can be confused because their finite sample properties are similar. In this paper we present a new approach to detecting multiple breaks in a series. The approach, which utilises the computational efficient methods based upon Atheoretical Regression Trees (ART), also allows categorisation of breaks as either 'spurious' or 'real'. We present empirical examples of the use of the approach utilising data on realised volatility from 16 Dow Jones Industrial Average index stock. Particular attention is placed on 5 stocks which exhibit long memory based upon the Beran (1992) test. Some statistical properties of the regimes identified are also considered.