The time-window-of-integration (TWIN) model (Diederich & Colonius, 2008) describes and predicts multisensory integration effects in (saccadic) reaction time as a function of the spatial, temporal, and intensity conditions of the stimulus set. Here we show that using identical stimulus conditions but different instructions about target/nontarget modality (redundant target vs. focused attention paradigm) TWIN makes different predictions that are supported empirically. Thus, TWIN is able to account for both bottom-up and tow-down effects.