One of the surest ways to spark a lively debate in a room full of coffee lovers is to ask a simple question: What is the best shape for a pour-over coffee filter?

Postdoctoral Fellow Dr. SCOTT FROST , Professors JEAN-XAVIER GUINARD and WILLIAM D. RISTENPART share early results from an ongoing research project conducted in partnership with SCA and Breville.

Anyone who has recently purchased pour-over equipment will be aware that modern coffee filters typically have either a “semi-conical” or “flat-bottomed” filter cup ( compare Figure 1 ). Fans of semi-conical filters tend to rave about the stronger flow of water through the grounds, greater extraction consistency, and correspondingly better flavor. Confusingly, fans of flat-bottom filters cite exactly the same benefits. Evidence supporting one view or the other tends to involve single-case reports of personal taste tests.

Image 1: Artist’s rendering of a flat-bottomed versus a semi-conical (“conical”) filter cup.

So far, we lack the hard data to answer this question scientifically and rigorously.

In 2017, SCA asked the University of California Davis Coffee Center to conduct research with the goal of expanding and updating the classic pour-over control charts for pour-over coffee. The research is still ongoing (with exciting initial results!), but as an initial "warm-up" project (and to get our own heated debates to a close), we decided to tackle the flat-bottom vs. semi-conical filter question. We were fortunate to work with Breville, who underwrote the entire project and provided us with their Precision Home coffee machine, which has easily interchangeable flat-bottom and semi-conical filters. The main idea was to carefully hold all other variables constant (roast level, water temperature, flow rate, etc.), and specifically evaluate the effect of filter geometry on the brewed coffee.

Designing the Experiment

Before we can answer the question, “Which shape is better?” we first have to answer the question, “Is there a difference?” It’s entirely possible that there isn’t actually any perceptible or noticeable difference between the two shapes, despite coffee lovers’ strong belief that there is. To determine if there is a difference, we first use “discrimination tests,” also known as “triangle tests.” In these tests, a subject drinks three cups of coffee, two of which are identical and one of which is different. The subject’s task is to identify the different cup of coffee. Since each subject has a one-third chance of guessing the correct cup of coffee simply by luck, we expect the subjects to correctly identify the different cup of coffee much better than 33%. If they can’t, there isn’t any noticeable difference.

We recruited 45 non-expert testers to come to the University of California, Davis Coffee Center and taste brewed coffee in sensory descriptive booths. These booths are very different from traditional tasting tables. They are designed to isolate the testers to minimize background bias and allow everyone to taste the coffee blind. The booths are also set to red light to minimize expectation bias (based on the black appearance that people think coffee should have).

Our triangle tests used a 2 x 2 factorial design, with two filter geometries and two different grind sizes (medium and medium-fine). This “factorial design” illustrates the importance of running six different triangle tests to cover all possible combinations of filter geometry and grind size (Table 1). We also ran a seventh triangle test, using a light roast versus a dark roast, as we used this significant flavor difference as an internal calibration and control. The focus here was on determining the impact of filter shape and grind size.

One of the most challenging things about running these experiments was nailing down the details of brewing coffee, because the logic of brewing three cups of hot, brewed coffee is important. For example, if two of the cups are hotter than the third, the test subjects will only perceive a temperature difference, not a taste difference, so we had to develop procedures to ensure that the temperatures of the coffees brewed using different methods were exactly the same when poured. Similarly, we found that we had to set a consistent "pour height" because some of our assistants tended to pour the coffee higher than others; this resulted in differences in the amount of milk foam on top of the coffee, and the test subjects could tell which cup was different just by visual observation. We spent a lot of time working out details like this to ensure the rigor of the experiments.

Calculation results

However, the work was worth it, as we obtained exciting and counterintuitive results, summarized in Table 1. Looking at the first row, we can see that 25 out of 45 participants correctly identified the different coffees when comparing the flat and conical filter shapes (both using medium particle size). Importantly, we expect one-third of the participants (15 people) to guess the result simply by chance, so the question becomes whether the 25 people getting it right is statistically significant or just a fluke. To answer this question, we use the “binomial probability distribution”, which describes the probability of multiple coin tosses (but in this experiment, the “coin” only comes up heads 33% of the time). Using this distribution, we calculate the probability (or “p-value”) that 25 participants guessed the correct coffee just by random chance. The p-value here is 0.004, which means that if we repeat the experiment 1,000 times, we expect to see 25 participants get the result right only in 4 of the trials. This is unlikely to happen! In other words, this is not a fluke, and there is a statistically significant and detectable difference between the two filters. We observed similar results when comparing flat and conical filter cups using fine grind sizes, with 23 of 25 participants identifying the correct cup of coffee (P-value 0.024).

Table 1: Comparison of drip filter base shapes and grind sizes.

So, for all the coffee lovers who insist that filter geometry affects flavor, our data shows that they are undoubtedly correct.

But surprisingly, our triangle tests also showed that non-expert testers were unable to tell the difference in grind size. Rows 3 and 4 of Table 1 show the results of triangle tests with the same filter geometry but different grind sizes. In the test with the flat filter, only 18 testers chose the correct cup; in the test with the cone, only 15 did. This result is not stronger than chance. The differences we found were not subtle: the median particle size for "medium" was 1065 microns, while the median particle size for "medium-fine" was 799 microns, a difference of 25%. For coffee lovers who often emphasize the extreme importance of precise control of grind size, it turns out that - at least for drip coffee - the average person can't tell the difference (at least in this grind size range).

The triangle test above demonstrated that there was a perceptible difference, but it didn’t tell us what the difference was. To address this question, we next conducted a series of detailed “sensory descriptive” tests. In these tests, we asked trained (expert) panelists to rate the intensity of various flavor attributes, such as “smoky,” “citrus,” “bitter,” and so on. We invited the best-performing panelists (i.e., those who correctly identified the three coffees the most times) to join our expert panel and trained them on the WCR tasting vocabulary with appropriate sensory references. We then had them rate 26 different flavor attributes, evaluating 8 different sample types in a “2 x 2 x 2” factorial design: 2 filter basket shapes, 2 grind sizes, and 2 roast levels. Our expert panel tasted each cup of coffee blindly in a sensory descriptive booth and provided a numerical ranking of the 26 attributes.

With 12 expert panelists each measuring 26 attributes and tasting eight different samples in triplicate, we had 12 × 26 × 8 × 3 = 7488 individual data points. Needless to say, describing such a large dataset is challenging. Figure 2 provides a graphical summary, showing “Principal Component Analysis” (PCA). This is an advanced numerical statistical technique that takes a large dataset and reduces it to the most meaningful “principal components”, helping to identify which groups are most dissimilar or most similar.

Image 2: The resulting principal component analysis (PCA) plot. The top plot brings together the most similar statistical results. We can see from this plot that the biggest differences occur between light and dark roasts, but the shape of the filter also produces a meaningful difference. The bottom plot summarizes the flavor attributes of each cluster. The top plot maps the bottom plot: light roasted coffee using a conical filter produces acid, citrus, and berry flavors; the filter produces raisin, dried fruit, and sweet floral flavors.

Several key pieces of information emerge from the PCA plot. First, as expected, there is a large difference between light and dark roasts, as evidenced by the large horizontal spread in both roast levels. More interestingly, we found that for a given roast level (such as the light roast on the left), the shape of the filter bowl also made a meaningful difference. The flat bottom filter bowl produced more dried fruit, sweet, and intense floral flavors, while the conical filter bowl produced more citrus, berry, and acid. The filter bowl shape had a similar effect on the dark roast, with the flat bottom producing more chocolate, cocoa, and wood flavors, while the conical filter bowl produced more intense bitterness.

Which is the best?

This is a key point, so let’s reiterate it: using the exact same coffee, the exact same water, the exact same temperature, and the exact same flow rate, we were able to change the flavor profile of the coffee in a very noticeable, statistically significant way simply by changing the geometry of the filter cup.

Having demonstrated that there is a difference, and what the difference is, the final question is, “Why is there a difference?” Filter geometry affects the way liquid flows across the base of the coffee grounds, thereby altering the “mass transfer” process by which molecules leave the solid coffee grounds and enter the liquid. The underlying details of this process are complex; this and many other aspects of brewing coffee are being actively and carefully studied at the University of California, Davis Coffee Center.

But for now, you might be wondering: Which filter shape is "best?" The answer, of course, is that the brewing method that produces the best tasting coffee for you is "best," and that's up to each individual to decide. But at least the next time someone starts arguing about the pros and cons of flat-bottomed vs. conical filters, you can use the hard data in this article to shed some light on the debate.

Professors WILLIAM D. RISTENPART and JEAN-XAVIER GUINARD are co-directors of the UC Davis Coffee Center, where Dr. SCOTT FROST recently received a postdoctoral fellowship.