How we rate water filters

The four inputs

Owner ratings, weighted by volume: 45 points

The star rating, mapped across a fixed window and then multiplied by a confidence factor based on how many people rated it. The window runs from 2.5 to 4.9 stars: 2.5 is the midpoint of the five-point scale, so a product at or below half marks earns nothing here, and 4.9 is close enough to perfect that almost nothing reaches it. Below 150 ratings the result is discounted in proportion; a product with almost none scores almost nothing here, however good the average looks.

Why it counts: A 4.8 from 12 buyers is a weaker signal than a 4.5 from 3,000. Weighting by volume stops a handful of reviews outranking years of evidence, and it is why a brand-new product will not lead one of our pages no matter how promising it is.

NSF/ANSI certification: 20 points

Whether an independent body has certified the filtration claims to a named standard: NSF/ANSI 42 for taste and chlorine, 53 for health contaminants, 58 for reverse osmosis. Scored in four tiers, because "certified to two standards", "certified to one", "tested" and "silent" are four different claims.

Why it counts: Any manufacturer can print "removes 99% of contaminants" on a box. Certification is the only part of that claim someone else has checked.

Filter cartridges: 20 points

How many filter cartridges the water actually passes through, measured against the number normal for its category. We count physical cartridges, not the "stage" number printed on the box.

Why it counts: A three-stage whole house system is normal; a three-stage reverse osmosis system is thin. The same number means different things, so it is judged per category rather than in the abstract.

Warranty: 15 points

Length of the manufacturer warranty, against the length standard for the category.

Why it counts: How long a manufacturer will stand behind a product is a reasonable proxy for how long they expect it to last.

How the certification tier is decided

This is the part of the formula we got wrong first, and it is worth explaining because almost every water filter site treats it as a yes/no. It is not. Our own research found that 50.1% of water filtration systems invoke NSF, ANSI, WQA or IAPMO while only 14.1% name a standard covering what they remove, so a binary check reads a great deal of unlabelled wording as certification.

The top two rows are the point of this whole page. They are both real certifications to real standards, and a listing describes them identically: "NSF/ANSI 58 certified". One of them means a laboratory checked the system for lead and fluoride. The other means it checked the system for dissolved solids. We score them six points apart, because for the person worried about what is in their water they are not the same product.

We do not count how many certificates a manufacturer holds, which was the obvious rule and the wrong one. It made a system certified for eight contaminants equal to one certified for dissolved solids, and it handed the top tier to products carrying NSF/ANSI 372, a lead-free plumbing standard that says nothing about filtration. Rewarding that would be rewarding the exact substitution our research is about.

The 0.25 tier is the other one that matters. Somebody did commission a test, so it is worth more than silence, and it is not a certification, so it is worth a great deal less than one. Every product card on this site states which tier it falls into, including when the answer is "none".

And we cite the source. Where a product is certified, the card names the certifying body, the exact model string that body's records use, which is usually not the name on the retail listing, plus the certificate number where one exists, the date we checked, and a link to the public record. It also states what the certification covers, because "NSF/ANSI 58 certified" can mean a system was checked for eight named contaminants or for dissolved solids alone, and the listing never says which. A verification you cannot repeat is just another badge, and this site exists because those are not worth much.

Those four figures are then divided by the best tier any product in the category actually reaches, and capped at 1, the normalisation explained in the next section. In a category where certification exists, a system whose certification reaches health contaminants earns the full 20 points, one certified only for dissolved solids or taste earns 14, and a merely-tested one earns 5. In a category where nothing is certified, "tested" is the ceiling and earns full marks, because a penalty applied equally to every option separates nothing.

The numbers in this table are read directly from the scoring code that produces every score on this site, so the table cannot drift out of date while the formula changes.

Why we count cartridges, not "stages"

The number of filtration stages a manufacturer advertises is not a comparable figure, and treating it as one quietly rewards whoever writes the boldest listing. The Waterdrop X8-WAN is sold as a "10-stage" system. The iSpring RO500AK is sold as "3-stage". Both are three-cartridge systems: a pre-filter, a reverse osmosis membrane, and a post-filter. Waterdrop counts the media layers inside each composite cartridge; iSpring counts the cartridges.

Scored as advertised, that difference was worth eight points on our own scale, awarded for a copywriting decision, on hardware that is substantially the same. So we count the cartridges the water passes through. It is the version of the number a reader can verify by opening the cabinet and counting, which is the test every input on this page has to pass.

Scored against the category, not against perfection

This is the part we got most wrong, twice, and it is worth explaining because it changes what a score means.

The first version scored against theoretical ranges: ratings out of five, stages out of six, warranty out of five years. It gave a well-reviewed system 48 out of 100. Everything landed between 40 and 50. The second version fixed the rating range but still measured certification against a perfect world, and rated the best whole house filter on the market 62.

A reader seeing 62 concludes "mediocre, keep looking". But 62 was not measuring the product. It was measuring the distance between that product and an imaginary certified one that does not exist in this category, and it applied the identical penalty to every candidate, so it did not even rank them. It was noise dressed as rigour.

A score answers "how good is this choice?", and a choice can only be made among things that exist. So each input is now scored against what is achievable in the category. Meet the category norm and you score full marks on that input; fall short and you lose points in proportion.

The benchmarks, published

Those last two columns come from our own survey of 441 systems, not from guesswork. Because no whole house system in that survey states a performance certification, certification cannot separate whole house products and is not allowed to drag them all down. The fact that the category lacks it is stated in prose on the page instead, which tells you something true. A uniform penalty told you nothing.

The consequence is worth being explicit about: a whole house system scoring 92 is not making the same claim as an under-sink system scoring 92. The first is excellent among uncertified options; the second cleared a bar that includes certification. Compare scores within a category, and read the certification line on every card, which is always stated, even when the answer is "not certified".

A score is not a league position

This is the most common way to misread the number, so it is worth stating plainly. The highest-scoring product on a page is not automatically the one you should buy. Our recommendation pages carry three to five picks that answer different questions, labelled on each card: best overall, best for well water, best tankless, lowest upfront cost. They are not competing for one spot.

A worked example. On the whole house page, the well-water system scores a few points below the city-water one, almost entirely because it has fewer owner ratings behind it. If your water leaves rust marks in the sink, the city-water system will not fix your problem at any score, and the well-water system is the only right answer on the page. The number tells you how good each system is against what its category can achieve. The label tells you which one is for you. Read the label first.

This is also why our cards carry no rank badge. They used to be numbered, and a "2" beside a product built for a job the "1" cannot do is a worse guide than no number at all.

What the bands mean

The rules the score is held to

• Every claim is checked at its source. A certification is looked up in the certifying body's own database by the manufacturer's model number, and the card names that database, the model string and the date we checked. Specifications come from the manufacturer's own documentation. Ratings come with the count behind them, so you can weigh them yourself. Nothing here is copied from a retail listing and repeated.
• Price is not an input. Prices move constantly, and a score that changes with a sale is measuring the sale, not the product. Value is a judgement we make in words, where we can explain it.
• Placement is not for sale. Brands cannot buy a score, a position, or a "best for" label. Scores are calculated by the formula above, so there is no lever to pull. We earn a commission when you buy through our links, it is disclosed on every page, and it is the same to us which product you choose.
• One product per review pool, per page. Amazon shows one review count across a family of variants, so a filter sold in four finishes reports the same total on all four. The figure we use is the one the listing shows a shopper, which means it can describe a product family rather than the exact model. We never place two products from the same pool on one page, because that would show you the same evidence twice wearing different names, and we discount a rating heavily when the count looks like it belongs to a newer product than the reviews behind it.

Where the data comes from

Two feeds, and it is worth being precise about which is which, because they do not always agree:

Ratings are the odd one out, and here is the honest reason. Amazon does not publish star ratings or review counts through its product API. The fields exist and a request for them is accepted; the response comes back empty, for every product we have ever asked about, including its own bestsellers. We retested this on 5 August 2026 to be certain it had not changed.

So the ratings on this site come from a third-party marketplace data feed instead, and that feed can lag the number printed on Amazon's page. On one product we currently list, the feed says 4.5 stars from 960 ratings and Amazon's own page says 4.3 from 613. We publish the feed's figure because it is the one that refreshes automatically and can be checked the same way every time, but you may notice the difference, and now you know why it is there. Where we have seen the two disagree it has moved a product's position by one place at most, never its presence on the page.

Every product card carries the date its data was last checked. If the price is more than three days stale the card stops showing it altogether and sends you to Amazon for the live figure, because a wrong price is worse than no price.

Where a manufacturer's claim cannot be verified against a certification or a published specification, we say so on the product rather than scoring it as though it were confirmed.