Some aspects of design of changeover experiments with autocorrelated observations are investigated. Wfiams' designs (1949, 1950) are seen to be balanced for autocorrelated observations. Acknowledgements. T h e author thanka the referee and an assodate editor for helpful comments and suggestions. -(1950). Experimental desigm balanced for pairs of residual dects. Austral. J. SU. Res. Ser. process. Austral. J. Statist. 26, 179-188. autoreppssive process. The Statist. 34, 161-173. A 3, 351-363.