When two targets (T1 & T2) are presented in rapid succession, observers often fail to report T2 if they attend to T1. The bottleneck theory proposes that this attentional blink (AB) is due to T1 occupying a slow processing stage when T2 is presented. Accordingly, if increasing T1 difficulty increases T1 processing time, this should cause a greater AB. The attention capture hypothesis suggests that T1 captures attention, which cannot be reallocated to T2 in time. Accordingly, if increasing T1 difficulty decreases T1 saliency, this should cause a smaller AB. In two experiments we find support for an attention capture hypothesis. In Experiment 1 we find that AB magnitude increases with T1 contrast – but only when T1 is unmasked. In Experiment 2 we add Gaussian noise to targets and vary T1 contrast but keep T1 ‘s SNR constant. Again we find that AB magnitude increases with T1 contrast.