A dual read-out model of context effects in letter perception is described that predicts forced-choice accuracy in the Reicher paradigm and its relation to word reportability. It is hypothesized that a correct choice to a letter in a word context is made when either the correct letter representation or a word representation containing the correct letter in the correct position reaches a response threshold (a criterion level of activation). This hypothesis was implemented using the basic architecture of the interactive activation model (J. L. McClelland & D. E. Rumelhart, 1981) in its semistochastic variant (A. M. Jacobs and J. Grainger, 1992). The model successfully captures the data of J. C. Johnston (1978), otherwise thought to be critically damaging for this type of model, and accurately predicts performance in a series of new experiments using the Reicher paradigm.