McConkie, Kerr, Reddix, & Zola [(1988). Vision Research, 28, 1107–1118] demonstrated that the distributions of landing sites on a word tended to be gaussian in shape. They provided a detailed account of the behaviour of the eye once a target had been selected and a saccade initiated, but said little about the process of target selection itself. The purpose of this study was to take as a starting point the landing site distributions of McConkie et al., in particular the residuals derived from fitting the gaussians to the empirical data, and to explore by computer simulation a number of saccade targeting strategies in order to discover candidates that best accounted for the residual data. Our results indicate that the strategy that gives the best fit involves targeting the longest word in a right parafoveal window extending 20 characters to the right of the currently fixated word. The implications of this finding for models of reading are discussed.