These data show that Piéron's Law holds for color saturation and under conditions of active response conflict, whereby two responses are both triggered by the same stimulus. These are both novel results. The latter demonstrates an extension of the relevance of this function beyond mere perceptual tasks and into the realm of cognitive tasks involving complex stimulus–response translation and active response conflict. The consistency of the results across the conventional and separated Stroop suggest that it is not due to some quirk of color information and word information being physically co-incident, as in the conventional Stroop, but is rather due to the nature of processing required by the task. The pattern of RTs in these experiments gives us some insight into the architecture of decision making. The additivity of the Stroop condition and color intensity factors, at first glance, seems to support the view that simple decision making can consist of two stages, detection and decision. This would certainly be the interpretation according to the AFM and is consonant with our analysis of the diffusion model style response mechanism. This analysis shows that a single-stage response mechanism which linearly combines stimulus evidence into a common metric cannot account for the pattern of data in tasks like this. This is important because the optimal integration of information over time and from different sources requires a common metric to represent the combined weight of evidence (Gold & Shadlen, 2002). Proofs of the optimality of diffusion model type evidence integrators (Bogacz et al., 2006) mean that alternative methods which are not formally equivalent must be non-optimal (according to this information integration definition of optimality). In the context of the current investigation, consider the scenario in which processing supporting responding to a stimulus was split into a detection stage, which recognized the elements of the stimulus display, and a decision stage, which reconciled these elements to calculate a response. In principle, this would mean that total decision time to barely detectable stimuli could be unfeasibly long (theoretically infinite, if it were not for noise in the decision process), even if the detection of such stimuli would require a response of the utmost urgency. In the current task, such a division would produce the lack of interaction between stimulus intensity and Stroop condition that we see here. This result is, of course, not proof that we cannot combine information from across stimulus dimensions to make a response. Rather, it is just a demonstration of a specific exception to the predicted RT performance from a naive informationally optimal model of evidence accumulation. Thus, this result fits within a wider body of work that shows that human decision making is not optimal (Kahneman, Slovic, & Tversky, 1982), when optimal is defined with respect to rational information integration. Previously, we have shown that models of optimal decision making may lead to ‘‘maladaptive’’ behavior when integrated into models of simple choice tasks (Stafford & Gurney, 2007). In particular, information optimality implemented by accumulation of evidence toward a threshold can lead to unrealistically long-selection times (if positive and negative evidence for a selection is closely matched), or selection of behaviors for which evidence is non-zero but negligible (if evidence accumulation is allowed to precede for a long enough time). Obviously, both of these outcomes would be undesirable for a real-world agent that must make rapid decisions, even between closely matched alternatives, and requires some minimal threshold of evidence for selection of behaviors. Here, we show that even simple choice tasks, such as the Stroop, can be made to reveal non-optimality in human decision making. A host of results show that the human and non-human primate brain can integrate information optimally, and in ways that can be modeled by simple choice models (Gold & Shadlen, 2007; Ratcliff et al., 2003). We suggest that although the primate brain can do this, it is not a complete explanation of primate choice behavior and its neural instantiation. If we expand our purview to include even marginally non-simple perceptual decisions, then the decision-making architecture reveals systematic non-optimalities. Our suspicion is that, as with many cognitive biases, the source of these non-optimalities will be the requirements of evolutionary adaptiveness (Gigerenzer & Todd, 1999). See Redgrave, Prescott, and Gurney (1999) for a discussion of this point specifically in the context of selection. These results also have an obvious interpretation according to the logic of the AFM (Donders, 1868–1869/1969; Sternberg, 1998). Although this approach is not currently fashionable, many authors do still infer multiple loci from the appearance of additive factors such as we found in our experiments (Pins & Bonnet, 1996; Woodman, Kang, Thompson, & Schall, 2008). Recent theoretical results have made it clear that the inference from additive factors to discrete processing stages is not trivial (Thomas, 2006; Townsend & Wenger, 2004), and our own work on processing stages in this version of the Stroop task confirms this (T. Stafford & K. Gurney, forthcoming). Whether or not it is possible to infer separate stages from additive factors, it is clear that, in our color saturation varying Stroop task at least, the underlying architecture for decision making appears to be able to resolve perceptual and response conflict in a way that keeps their effects on decision times separate. The implication of this is that although great progress has been made in modeling simple perceptual decisions, and although pervasive regularities, such as Piéron's Law, may hold in more complex decisions, the generalization of these models to marginally more complex decisions is likely to be far from straightforward.