Arguably, the first person to attempt an answer to this question was English psychologist Edward Titchener, who in 1896 concluded that a human is capable of distinguishing 33,000 colors [ ].

In 1939, Edwin Boring, a professor of psychology at Clark University and Harvard University, cited a value of 300,000 colors [ ]. That same year, American physicist Dean Judd—who made significant contributions to scientific colorimetry—estimated that a trained observer could distinguish approximately 10,000,000 colors [ ].

In 1943, American researchers Dorothy Nickerson and Sidney Newhall determined that under ideal viewing conditions, a person could distinguish 7,295,000 colors [ ].

Presumably, such a wide range of values arises due to two main factors: on one hand, the viewing conditions at the time were not precisely defined; on the other hand, there was a lack of accurate measurement instruments for standardizing the set of reference samples.

To estimate the number of perceivable colors, scientists over time have used roughly the same methodology. It is based on the idea that the number of color gradations a person can perceive is limited by the resolving power of the human visual system.

In the physical world, the stimuli that trigger a color response in the “eye-brain” system are continuous—that is, they change infinitely smoothly and are not made up of discrete steps. However, the tools used to measure these stimuli operate on a discrete basis, which allows us to numerically define the limits of perceptual capability.

For example, if we divide a certain color range into a number of gradations (or shades), increasing the number of these gradations will eventually lead to a point where an observer can no longer distinguish the difference between adjacent ones.

In other words, it is possible to instrumentally measure the number of gradations at which a change in color becomes so smooth for the observer that two adjacent shades appear identical. Any further increase in the number of shades within the considered range will no longer lead to an increase in the number of distinguishable colors.

Based on such data, one can calculate the maximum set of colors that a human can distinguish. For example, if we assume (or obtain experimentally) that a person can distinguish, say, 200 shades along the hue circle, 450 levels of lightness, and 80 levels of saturation, then we arrive at a value of 200 × 450 × 80 = 7.2 million colors (this number is given purely as an illustrative example).

Obviously, the real formula is more complex, because the human ability to distinguish colors varies across different ranges of hue, lightness, and brightness. Moreover, it is essential to clearly define the observation conditions, as the results may vary significantly depending on these factors.

It is likely that when Titchener conducted his research in 1896, his technical means for accurately measuring closely related shades of color were quite limited. At that time, computers, monitors, colorimeters, and spectrophotometers did not yet exist. Therefore, the results he obtained are unlikely to be considered accurate by modern standards.

Over time, both research capabilities and knowledge of color perception advanced, leading to new estimates: around 300,000 colors (1939), approximately 10,000,000 colors (also 1939), and finally the relatively precise figure of 7,295,000 colors in 1943.

Scientific articles published since then up to the present day still refer to those earlier data. However, we were unable to find any new studies, despite the fact that modern color reproduction tools clearly allow for the display (and precise measurement) of color stimuli with far finer gradation steps than those available in 1943. That is why the Dehancer team decided to conduct its own experiment.

In April 2023, we released the app for iPhone, which allows each user to determine the number of perceivable colors based on perceptual error across various ranges of hue, brightness, and tone within the 24-bit RGB color model—that is, within a space of 16,777,216 colors.

Since the testing methodology is based on measuring color matching error (rather than the total visible color range), we believe that factors such as screen gamut (color coverage), colorimetric calibration accuracy, and display shifts (e.g., True Tone mode) do not significantly affect the test’s accuracy. The technical precision of the result is primarily determined by the minimum gradation step that the device is capable of reproducing. In our case, the testing is conducted on iPhones and iPads—overwhelmingly newer generations with Super Retina XDR displays. These devices provide sufficient color resolution and therefore fully meet the requirements for this experiment.

The maximum number of color gradations that can be reproduced is one of the experiment’s limitations. In other words, we are not measuring how accurately a given device reproduces color—we are measuring the user’s ability to distinguish between colors. According to our hypothesis, if we observe a significant number of users reaching the upper measurable limit of the color model, it would indicate that the RGB color space used in the test is insufficient for this type of experiment.

We divided user testing into two alternating task types:

According to our hypothesis, the first case measures the physiological limits of the user’s vision, determined by the sensory system of the eye. For this, we use the most ideal conditions possible within the constraints of our model: the color patches are displayed as one square nested tightly within another, allowing direct and precise visual comparison.

In the second case, perceptual color perception is measured, involving short-term memory. For this, the color patches are displayed at a certain distance from each other. With some caveats, this test reflects the ability to perceive colors during simultaneous viewing. This particular metric is of primary interest within the context of our research.

Over the first 10 months, more than 7,000 people participated in the testing, completing a total of 10,400 tests. Out of those, 9,620 were statistically valid and suitable for analysis. We filtered out tests from users who indicated an age over 130, those who completed the test in under 2.5 minutes (typically users who simply clicked “Next” without attempting to match the colors), as well as other clearly invalid data entries.

After gathering and analyzing the initial dataset, we are ready to share some preliminary results.

The average result in the sensory perception test was approximately 10,095,000 colors. The maximum was 14,616,000 colors. Based on this data, we can state that at the sensory perception level:

None of the participants reached the theoretical maximum of 16,777,216 colors within the 8-bit RGB model in the sensory test. The best result—14,616,000 colors—was achieved by only one user (0.01% of all tests). Thus, the 8-bit RGB model can be considered sufficient both for conducting such research and for reproducing colors in everyday life.

Since the primary focus of our research is perceptual color perception, once we had gathered enough data to answer the main questions regarding the sensory test, we released a new version of the app. In this version, we removed the sensory test and slightly expanded the perceptual one (increasing the number of color patches from 15 to 24). After this update, we continued collecting data for another 5 months.

The average result in the perceptual perception test was approximately 4,200,000 colors. The maximum was 13,200,000 colors (achieved in a single test, which accounts for about 0.01% of all tests taken). Based on this data, we can state that at the perceptual level of color perception:

The chart shows how the number of perceived colors changes depending on the number of attempts. The central line represents the average result, while the blue band indicates the confidence interval, within which the theoretical mean is expected to fall with 95% probability according to probability theory.

The graph clearly demonstrates that as the number of test attempts increases, the average number of distinguishable colors shows a consistent upward trend—up to the 5th attempt. For a higher number of attempts, there is currently insufficient data for meaningful analysis, as there are too few such users to form a statistically significant sample.

Based on this data, we can confidently state that repeating the test leads to improved results. On one hand, this may suggest that color perception can be trained. On the other hand, the exact reasons for this improvement remain unclear. It could be due to increased attention, higher standards for matching colors, longer test durations, or other factors.

The average time to complete the test is approximately 7 minutes, with the majority of users completing it in around 4.5 minutes.

The chart shows that, on average, taking more time to complete the test tends to result in a higher score, but this correlation is relatively weak, as there is less data and greater variability for longer test durations.

It can be assumed that results improve with longer test durations because this allows users to pay more attention to each step (i.e., each color patch in the test) and/or make use of technical tricks. For example, by shifting their gaze between the iPhone screen and elsewhere, users can reset residual retinal stimulation, effectively “zeroing out” the visual system’s adaptation and thereby temporarily increasing sensitivity to subtle color differences.

In addition, having more time enables the use of other “hacks” that interfere with the adaptation of the visual system — such as alternating between using the left and right eye, changing the viewing distance by moving the phone closer or farther away, tilting or rotating the phone, and so on.

If we divide the test completion times into equal intervals of approximately 3 minutes each, we can see that as the duration increases — specifically, in the range from about 322 seconds (5:25 minutes) to 1182 seconds (19:45 minutes) — the average number of perceived colors consistently increases, eventually growing by about 20%, from approximately 4,200,000 to 5,100,000. At this time, there is insufficient data to analyze test results with longer durations.

The graph shows that color perception tends to improve slightly with age on average, although this correlation is weakly expressed. The theoretically calculated regression line is clearly visible in the older age range (right side of the graph), where the amount of data is the smallest.

Breaking down the test results by age groups of approximately 12 years each shows that in the range from approximately 16 to 60 years old, the average number of perceived colors steadily increases, ultimately growing by 34% — from around 3,600,000 to 4,800,000 colors. The amount of data for older age groups drops sharply, so at this time it is difficult to draw reliable conclusions about color perception in the elderly.

To better understand the factors behind this age-related trend, further data analysis is needed — one that simultaneously accounts for both age and test completion time.

If we divide all participants into four age groups with an equal number of test results, and within each age group further divide the results into four subgroups based on test duration — again with an equal number of tests per subgroup — we arrive at the following chart.

The graph shows that at the same test duration, older users tend to achieve slightly better results than younger ones. In the first three time-based subgroups, this trend is clearly visible. In the last subgroup, it is less pronounced due to a large variance in the data, which currently makes it difficult to draw any reliable conclusions for the longest test durations.

This suggests that, all other factors being equal, people perceive colors slightly better as they age. Moreover, since this trend is not related to test duration, it is presumably not linked to factors like concentration, diligence, or perseverance.

We thank everyone who took the test — especially those who did it more than once. We would also greatly appreciate it if anyone is willing and able to take the test a few more times — this will provide us with more data for further analysis. Please note that we do not collect or store any personal data: all statistics we gather are completely anonymized.

[1] E. B. Titchener, Oulline of Psychology, Macmillan, New York, 1896

[2]   

[3] D. B. Judd and K. L. Kelly, «Method of designating colors» J. Research Nat. Bur. Stand. 23, 355, 1939

[4]   

[5]