When designing an experiment, one issue that every researcher must deal with is whether to present each condition of the experiment in a separate block of trials (a pure block design) or whether to combine all the conditions in a single block of trials (a mixed block design). The choice can have considerable consequences for the ultimate interpretation of the results. That is, pure blocks provide the optimal opportunity for the use of condition-specific strategies, which can lead, at the very least, to quite different levels of performance in the pure versus mixed block situations. A common finding from mixed block designs is that there is a mixing cost (Los, 1996). Each condition has longer latencies than when those same conditions are presented in pure blocks. In word/nonword naming tasks, however, a different, homogenization, pattern emerges. While there is a mixing cost for the easier stimuli, there is a mixing benefit for the harder stimuli (Lupker, Brown & Colombo, 1997). The present talk will discuss the generality of the homogenization pattern and its implications for models of word recognition. A number of possible explanations of these types of results will be considered, including the pathway-selection/route-emphasis account (Monsell, Patterson, Graham, Hughes & Milroy, 1992; Zevin & Balota, 2000), the input gain hypothesis (Kello & Plaut, 2000) and our own, time-criterion, account (Lupker et al., 1997).