Abstract Virzi and Egeth (1985) have proposed a model of Stroop interference in which interference is attributed to the translation of information from one processing system to another. They used a task that should not require translation: manually signalling the position of a conflicting position word or vocally reading the word. Because little interference occurred, they concluded that translation is not obligatory. The present Experiment 1 showed that substantial interference occurs when participants must be prepared to respond to either dimension, suggesting that translation is an obligatory process when attention is divided between the two dimensions. Experiment 2 showed that this interference is reduced if a cue that indicates the relevant dimension is presented 300 or 1000 ms prior to the stimulus. These results are discussed in terms of a strength of processing model in which the degree of Stroop interference is attributed to the degree of attention to the irrelevant dimension of the Stroop stimulus. This model assumes that the degree of attention to a stimulus dimension is subject to both top-down and bottom-up influences. In most versions of the Stroop task, participants are presented with a word and are instructed to ignore the meaning of the word and to name some other attribute of it, such as its color or position. When the word conflicts with the attribute to be named, interference occurs in that responding is slowed compared to naming the same attribute of a nonword stimulus. For example, if the word left is presented to the right of a fixation point and the task is to name the position of the word, reaction times (RTs) are longer than when naming the position of a string of asterisks or Xs. On the other hand, if the task is to read the word aloud, the color or position of the word has little, if any, on RTs. The absence of what is usually referred to as a Stroop effect reflects an asymmetry in the processing of information about the various dimensions of the Stroop stimulus: Word reading interferes with color or position naming but not vice versa. Various models have been proposed to account for this asymmetry in information processing. The oldest and, until recently, the most popular explanation for this asymmetry was in terms of the relative speed of processing of information about the dimensions of the Stroop stimulus (e.g., Stroop, 1935; Dyer, 1973). This view assumes that information about the two dimensions is processed in parallel, but that these parallel pathways converge on a common output mechanism. Word reading is assumed to be faster than position or color naming, with the consequence that information about the word gains access to response mechanisms earlier than information about color or position, thereby generating Stroop interference. However, it has been found that presenting information about the slower dimension prior to the word stimulus does not produce reverse Stroop interference when the task is word reading (Glaser & Glaser, 1982). Similarly, presenting the word prior to the other dimension does not increase interference when the task is to respond to that other dimension (Glaser & Dungelhoff, 1984). MacLeod and Dunbar (1988) used a Stroop-like task where participants first learned associations between color names and novel shapes. After 1 to 2 hours' practice in shape naming, colors were named much faster than shapes; after 5 hours of practice RTs for shape naming had decreased greatly, and after 20 hours of practice RTs were about the same. The patterns of Stroop and reverse Stroop interference, found in these different conditions, however, were inconsistent with a speed of processing account. Virzi and Egeth (1985) have proposed a translational model as an alternative to the relative speed of processing model. They propose that there are several systems which process information about the Stroop stimulus in parallel, and that each of these systems codes and processes information in a way that is specific to that system. …