Every year, people eagerly await the arrival of the fall colors, and if you’re like me, you’ve checked the forecasts for peak foliage to find the best time to go leaf peeping. You might be surprised to learn that there’s a fair bit of science that goes into forecasting the arrival of those fall leaves, based on factors like the number of daylight hours, changes in the temperature, and the amount of rainfall. Scientists have even developed machine-learning models for predicting when the leaves change. But that’s not the only science that goes into appreciating the fall foliage: there’s also a lot of cognitive science that aims to understand how exactly we perceive these amazing colors.

For example, if you take a look at the picture on the left below, you can easily see that those leaves are just starting to change color, compared to the ones in the picture next to it. To do this, you don’t need to inspect every single leaf on those trees (that would be incredibly time-consuming!). Instead, you can quickly get an impression or an overall sense of their typical color, without needing to focus on the individual items. The visual process underlying these judgments is called ensemble perception. And it’s not just for fall leaves—this is an essential part of much of our visual experience. If you look at the pictures on the right, you can easily judge which way the people are generally headed as they cross the intersection, or the typical size of the buttons. We can efficiently form ensemble judgments about many features, including color, motion, size, and facial expression.

But exactly what kind of information do we get from these ensembles? For example, are we representing some information about the overall average, the range of colors, or are we getting something else? In other words, what kind of information can we consciously access from a set of objects? This is the question that Vladislav Khvostov, Árni Gunnar Ásgeirsson, and Árni Kristjánsson (pictured below) investigated in a recent paper in the Psychonomic Society journal Psychonomic Bulletin & Review.

The authors set out to test between a couple of possibilities. The first is that our representation might be limited to a couple of summary statistics: information about the average as well as the spread of the items (i.e., what is the average color of the leaves, and how much do they vary?). The other is that we might have a richer representation of the items and know something about how the features are distributed within the set. For example, we might be able to say that there are plenty of greens and reds in a set of leaves, but fewer yellows.

To examine this question, the authors developed a new experimental task for probing what information participants are able to consciously access. Participants viewed displays like the ones shown in the second panel below. After each display, participants were shown a random probe color (e.g., the green dot in the third panel), and were asked to rate how frequent the probe dot was in the display they just saw.

Importantly, the authors varied the color distributions of the dots in each display. In one condition (shown in the left panel below), the distribution of dots was based on a bell curve (Gaussian) distribution–dots that corresponded to the average color were more prevalent, compared to colors that were far from the average. In another condition (shown in the middle panel), the frequency of each color was the same across the entire range (uniform distribution). Finally, a third condition (right panel) used a bimodal distribution, such that the dots clustered around one of two colors (for example, yellow and blue). Importantly, the average and the overall range of the colors were exactly the same across each of the three conditions, and only the distributions of the colors varied.

The authors reasoned that if all we have access to is information about the average color and the range of colors, participants’ reports of the prevalence of the probe colors should be the same across all three conditions, and they would report the average color as being the most prevalent in all cases. On the other hand, if participants know something about the distributions of the colored dots, their reports should match the actual distributions of the colors shown.  As shown in the graphs below, that is exactly what the authors found. Participants were remarkably accurate in reporting the prevalence in each condition.

Together, these results show that our perception of groups of objects is richer than previously believed. When we get an overall sense or impression of those lovely fall colors, we’re taking in more information than you might think! We’re not just registering information about the average color (yellow) or the range of colors (red to green). Instead, we have a rich representation of how those colors are distributed—whether they’re a mix of red and green, mostly yellow with relatively few greens, and so on. This may also help explain why we have a sense that we’re taking in a lot of visual information at any given moment. We know that we aren’t inspecting every leaf (unless you’re a particularly enthusiastic leaf peeper!). Instead, we rely on a highly efficient ensemble perception process to support our conscious visual experience.

Psychonomic Society article featured in this post

Khvostov, V., Ásgeirsson, Á. G., & Kristjánsson, Á. (2025). Explicit access to detailed representations of feature distributions. Psychonomic Bulletin & Review, 1-10. https://doi.org/10.3758/s13423-025-02716-3