When two targets are embedded in a temporal stream of distractors, second-target identification is initially impaired and then gradually improves as intertarget interval lengthens (attentional blink; AB). According to bottleneck models of the AB, difficulty of first-target processing should modulate the magnitude of the second-target deficit. To test this, we examined whether a data-limited manipulation of T1 difficulty (forward masking) would modulate AB magnitude. In two experiments, we show that data-limited manipulations of T1 difficulty do affect the AB, so long as T1 is not masked by an immediately trailing distractor. When such a trailing item is present, the relationship between T1 difficulty and the AB disappears.