How We Rate Mattresses: The Research Score, Explained

Every mattress on OpenMattressLab gets a single Research Score out of 10. It is calculated, not decided. Half of it comes from how the mattress performed in published lab-style testing; three-tenths from what independent expert reviewers concluded; one-fifth from what owners say on review platforms. The code that produces the number is the same code that renders this page, so every figure below is computed live from our data — if the formula changes, this page changes with it.

We built it this way because each of the three evidence types fails in a different way, and leaning on any one of them alone produces bad recommendations. Lab measurements are objective, but they come from one sample of one mattress, tested for weeks rather than years. Expert reviewers sleep on more beds than anyone, but they sometimes have commercial relationships with the brands they review, and their scales vary wildly. Owner reviews capture years of real use, but they are self-selected, easily curated by brands, and often about delivery rather than comfort. Blending all three, with explicit weights, lets the strengths of each cover the weaknesses of the others.

What the score is not: a prediction that you personally will love a mattress. Firmness preference, body weight and sleep position matter more to your comfort than any overall rating. A 9.0 mattress that is too soft for you is worse than a 7.8 that fits. That is why every ranking on the site pairs the score with “best for” and “not for” lists, and why the second half of this page explains how to choose a firmness. Use the score to build a shortlist, then use fit to choose.

Performance is the largest slice because it is the most objective. Each mattress gets seven sub-scores from 0 to 10, which are combined with fixed weights into a single performance number. Here is the split:

• Pressure relief (20%) — how well the comfort layers cushion shoulders and hips. It is the main driver of comfort for side sleepers and anyone with joint pain.
• Spinal support (20%) — whether the mattress keeps the spine in a neutral line, especially under the heavier hips. Poor support is the most common cause of waking with a stiff back.
• Durability (13%) — expected longevity, based on materials and foam densities. A mattress that sags in three years is bad value however good it feels on day one.
• Value (15%) — performance per dollar at the typical (not list) queen price.
• Cooling (12%) — how much heat the surface traps. Foam-heavy beds run warmer; coils and latex breathe.
• Motion isolation (10%) — how much a partner’s movement travels across the bed.
• Edge support (10%) — how much the perimeter collapses when you sit or lie near it, which effectively changes the usable width of the bed.

Why these weights? Pressure relief and spinal support together make up 40% because they determine whether you sleep comfortably and wake without pain, for every sleeper, every night. Durability and value come next because a mattress is a seven-to-ten-year purchase; a bed that loses its support early, or costs twice as much as an equivalent, should not rank at the top. Cooling, motion isolation and edge support matter a great deal to some people and barely at all to others — a single sleeper doesn’t care about motion transfer — so they carry smaller weights in the general score and get their own dedicated rankings, such as our cooling and couples guides, where they matter most.

We don’t operate a mattress lab, and we never claim to have slept on the beds we rate. Every sub-score is taken from, or derived from, someone else’s published testing — and we follow a strict order of preference.

First choice: RTINGS’ measured scores. RTINGS is the most transparent mattress tester we know of. It runs the same instrumented tests on every mattress — firmness measured in pascals per millimetre, pressure mapping, and repeatable heat, motion and edge tests — and publishes its scores and test methods. When RTINGS has tested a model, we use its numbers for pressure relief, support, cooling, motion isolation, edge support and longevity directly, without adjustment.

Second choice: Sleep Foundation and Sleepopolis, minus about one point. Many mattresses were never tested by RTINGS. For those, we use the category scores published by Sleep Foundation and Sleepopolis, which rate the same attributes. But their scales are more generous than RTINGS’ stricter, measurement-based scale. So we subtract roughly one point from those sub-scores to put them on the same footing. Without that correction, an untested mattress would look better than a tested one simply because the reviewer was kinder — a bias we refuse to build in.

Value is our judgment, and we say so. No lab measures value. We set the value sub-score by comparing a mattress’s measured performance with its typical selling price — not its inflated “list” price, because almost every online brand runs a permanent sale. A bed like the DreamCloud Classic Hybrid, which scores well at around $699 for a queen, gets 9.5; luxury beds with similar performance at three times the price score lower. It is the one place editorial judgment enters the performance slice, and it is capped at 15% of it.

Every model page lists the exact RTINGS, Sleep Foundation or Sleepopolis review behind its numbers, so you can check any sub-score against its source. If you see one that looks wrong, the about page explains how to tell us.

Aggregating evidence would be easy if every source agreed. They often don’t. In our October 2026 data review we found conflicts between RTINGS, Sleep Foundation and Mattress Clarity on many of the models we cover, and most of them fell into three patterns. Here is what we do with each, so you can predict how any number on the site was chosen.

Firmness labels versus measured firmness. A brand calls a bed medium-firm; one reviewer rates it 6.5 out of 10 by feel; RTINGS measures it in pascals per millimetre and finds it firmer or softer than either. When a measured value exists, it wins, because it was taken the same way on every mattress. We still show the reviewer’s feel-based rating next to it, since a heavier tester and a lighter one genuinely experience the same bed differently — that disagreement is information, not noise. Where no measurement exists, we report the range across reviewers rather than picking the number we like best.

Weight-split scores versus one overall number. Sleep Foundation and some other outlets rate a mattress separately for lighter, average and heavier sleepers, while others publish a single overall score. These are not the same quantity, so we don’t average a weight-split rating with an overall one. The expert slice uses each outlet’s overall score. The weight-split ratings go into our “best for / not for” lists and our heavy-sleeper and side-sleeper guides, where they are the most useful evidence there is.

Outlets testing an older version. Brands revise foams, coils and covers without renaming the model, so a review from two or three years ago may describe a different mattress. When a test predates a known construction change, we label it with the version tested, prefer a newer test of the current build where one exists, and drop the old figure from the score if the change plausibly affects it. If only an old test exists, we keep it and say so on the page.

The expert slice captures something lab numbers miss: the overall judgment of people who have spent nights on hundreds of mattresses. We record each independent outlet’s verdict — Tom’s Guide, Sleep Foundation, Sleepopolis, Mattress Clarity, Forbes Vetted, NapLab, RTINGS and others — and, where the outlet publishes an overall score, we normalise it to a 10-point scale (a 4.6 out of 5 becomes 9.2).

The expert score is then the average of those numeric scores, plus 0.1 for every award the mattress holds in a published best-of guide, counting at most five awards, and capped at 10. Awards matter because they show an outlet chose this bed over its competitors for a specific job, not just that it rated it well in isolation. Capping the bonus at five awards (+0.5) stops heavily marketed beds from stacking dozens of minor “best for” badges into a perfect score.

Two rules keep this honest. Outlets that only publish a written verdict, with no number, don’t enter the average — but their awards still count. And if a mattress has no numeric expert scores at all, we use a neutral 8 rather than guessing. In practice, almost every mattress we cover has at least two scored reviews, so that fallback rarely applies.

Why only 30%? Expert reviewers are skilled, but they are also a small sample of bodies — often one or two testers — and they tend to review a mattress soon after launch, before durability problems surface. Their verdicts are excellent at telling you how a bed feels and who it suits, which is why we quote them throughout the site, but they shouldn’t outweigh instrumented measurements taken the same way on every mattress.

We know many of these outlets earn affiliate commissions, as we do. That is exactly why the expert slice is 30% and not more, and why we average across several outlets rather than trusting any one. When a reviewer has a direct ownership conflict — such as a review site owned by a mattress company — we say so on the page where it matters, as we do for Douglas in our Canadian guide.

Owner reviews are the only evidence that covers years of use, so they belong in the score. But they are also the easiest to game. A brand can display only its best reviews, prompt happy customers at the moment of delivery, or pool reviews across every product it sells. So we weight owner ratings with three deliberate safeguards.

1. Square-root weighting by review count. Each platform’s influence grows with the square root of its review count, not the count itself. A platform with 10,000 reviews counts ten times as much as one with 100, not a hundred times. That stops a single enormous widget drowning out a smaller independent sample.
2. Brand-run widgets count a quarter. Ratings shown on a brand’s own website — which nearly always sit at 4.7 to 4.9 stars — are multiplied by 0.25 before weighting. Independent platforms such as Trustpilot, GoodBed, the BBB or Amazon count in full.
3. A hard cap of 9.6. However glowing the reviews, the owner component can’t exceed 9.6 out of 10. A 4.9-star average on a brand site is a marketing number; it shouldn’t be able to lift a mattress above one with better lab results.

The Saatva HD shows the cap in action. Its only owner rating is Saatva’s own site, at 4.9 stars from 307 reviews. That converts to 9.8 out of 10, which the cap trims to 9.6. With only one platform the 0.25 multiplier doesn’t change its share — but the moment an independent platform appears, as it does for the DreamCloud example below, the brand widget’s influence shrinks sharply. Brand-level Trustpilot and BBB records aren’t mixed into model scores (they mostly measure delivery and service, and samples are small); we show them prominently on every brand review instead, and flag any score that rests on brand-run reviews alone with a † in our comparison tables. If a mattress has no owner ratings at all, it gets a neutral 7.5.

We treat a brand-run widget as evidence that owners exist and are broadly satisfied, not as a measure of how satisfied. The brand chooses when to ask for a review, which products share a pool, and how disputes are moderated. Independent platforms have their own flaws — Trustpilot skews to delivery and service stories — but the brand can’t curate them, which is why they count four times as much.

Here is the whole calculation for one real mattress, step by step. We chose the DreamCloud Classic Hybrid because it has every type of input — RTINGS lab data, several scored expert reviews, an award, and both a brand-run and an independent owner rating. Every number in the tables below is computed from our live data with the same functions that score the rest of the site.

Step 1 — performance

RTINGS tested this model, so its measured scores are used directly. Value is our judgment at the typical queen price of $699.

The standout numbers are edge support (9.0) and value (9.5); the weak spot is durability (5.5), which RTINGS attributes to lower-density foams. Weighted together, performance comes to 7.69.

Step 2 — expert consensus

4 outlets published a numeric score, averaging 8.88. One award adds 0.1. Note that RTINGS appears in the list without a score: its overall verdict is captured in the performance slice already, so counting it twice would be double-dipping. The expert score is 8.97.

Step 3 — owner ratings

This is where the safeguards earn their keep. The brand’s own site shows 4.8 stars from 9,383 reviews; the independent platforms tell a harsher story. Counted naively by volume, the brand site would dominate. After square-root weighting and the brand-run discount, independent platforms carry 61% of the weight, and the owner score lands at 8.63 — well below the brand’s headline 4.8 stars, and a more honest reflection of what owners report.

Step 4 — the Research Score

Put together, the DreamCloud Classic Hybrid scores 8.3. You can see it in context in our best mattress ranking and our DreamCloud review, where we explain who it suits (back sleepers, couples, edge sitters) and who should skip it (sleepers over 230 lb, light side sleepers who want to sink in).

Because the formula is fixed, you can compare any two mattresses on the site directly. Here is every model in our database, scored by the same code. If a number here ever disagrees with a review page, that is a bug, and we want to know.

Notice how compressed the range is. Most mattresses that make it into serious expert coverage are decent; the difference between a 7.8 and an 8.3 is real but modest, and it is smaller than the difference a wrong firmness makes to your sleep. Treat anything within about 0.3 points as a tie and decide on fit, price and policies instead. The bottom of the table is more meaningful: a score in the low 7s usually reflects a specific, repeated weakness — weak edges, heat, short lifespan — that the review page spells out.

The Research Score tells you how good a mattress is. Firmness tells you whether it is good for you. These are the rules we use throughout the site, including in our 30-second mattress finder, and they are grounded in the best available sleep research.

Start with your sleeping position. Side sleepers put their weight through the shoulder and hip, so they need a softer surface — around 5 out of 10 — that lets those points sink while the waist stays supported. Back sleepers need more support under the hips to keep the lower back from arching, so medium-firm, about 6.5, suits most. Stomach sleepers need the firmest surface, around 7.5, because a sagging midsection bends the spine backwards. Combination sleepers do best around 6: firm enough for back and stomach, soft enough for side, with enough bounce to turn easily.

Then adjust for body weight. The same mattress feels firmer to a lighter person and softer to a heavier one, because a heavier body compresses the comfort layers further and reaches the support core. That is why we suggest roughly one point softer for sleepers under about 130 lb and one point firmer for those over about 230 lb. Heavier sleepers should also prefer hybrids, whose coils resist bottoming out better than all-foam beds; our guide for heavier people covers this in depth. Sleep Foundation’s firmness guidance follows the same logic.

For back pain, medium-firm beats firm. The most-cited trial on the question, Kovacs and colleagues’ 2003 randomised study in The Lancet, compared firm and medium-firm mattresses in people with chronic low-back pain and found the medium-firm group reported better outcomes for pain in bed, pain on rising and disability. A 2015 systematic review by Radwan and colleagues in Sleep Health concluded that medium-firm mattresses, along with custom-inflated designs, best promoted sleep comfort, quality and spinal alignment. Caggiari and colleagues’ 2021 review reached a similar conclusion for low-back pain, and Jacobson’s 2009 study found that replacing old bedding with new, sleep-position-appropriate bedding improved reported sleep quality and back discomfort. That is why our firmness rule nudges back-pain sufferers only half a point firmer, not into “orthopaedic” territory — see our back pain guide for the picks.

Treat brand firmness labels with suspicion. “Medium-firm” means different things at different companies. Wherever possible we report firmness as testers measured it rather than as the brand describes it — and where testers disagree, as they often do, we show the range. The DreamCloud Classic Hybrid is a good example: its firmness note reads “Testers 6–7.5; brand now labels it Firm; RTINGS measured Medium-Plush (39 Pa/mm).”

Trust comes from being clear about limits. Here is what this site does not do, and will not pretend to do:

• We don’t run an in-house lab. We don’t own pressure mats or thermal cameras. Lab data comes from RTINGS and other named testers, and we credit them on every page.
• We don’t claim to have slept on the mattresses. You will never read “our testers loved it” here. We say who measured what.
• We don’t invent quotes or numbers. Every owner quote links to where it was published, and every number traces to a source in the list at the bottom of the page.
• We don’t accept payment for placement. A brand can’t pay to be ranked higher, and partner brands get no adjustment in the formula.
• We don’t use brand-supplied review scores at face value. They are down-weighted, as described above.
• We don’t score regional mattresses with the US formula. RTINGS rarely tests UK, Australian or Canadian models, so our UK, Australia and Canada guides rank on expert agreement and owner service records instead, and say so.

It also means we can be wrong in the same places our sources are wrong. If RTINGS received an unusual sample, or a reviewer had an off week, that error flows into our numbers. We reduce the risk by preferring measured data, by averaging several expert outlets instead of trusting one, and by keeping owner reviews in the mix as a long-term reality check. When sources disagree sharply — a mattress one outlet calls soft and another calls firm — we don’t paper over it; we report the range and tell you which kind of sleeper each verdict seems to fit.

We think this is more honest than the alternative. Many mattress sites describe “testing” that amounts to a few nights on a free sample sent by the brand. Aggregating several independent sources, and showing the working, gives you a sturdier answer than any single person’s impression — and lets you check us.

OpenMattressLab earns commissions when you buy through some of our links — through Amazon Associates in the US, UK, Canada and Australia, and through affiliate programmes for Saatva, Nectar, DreamCloud, Puffy and Bear. The full list is on our disclosure page. Buy links are marked rel="sponsored".

The formula is our main protection against that conflict. Because the score is computed from published data with fixed weights, there is no lever for a commission to pull. Partner and non-partner brands go through identical code. We routinely recommend beds that pay us nothing when they’re the better fit, and we publish complaints about partner brands as prominently as praise — for example, the low Trustpilot scores of some partner brands sit right next to their buy buttons.

The same rules apply to brands we have no commercial relationship with. We now publish full reviews of Purple, Helix, Casper, Tempur-Pedic, Beautyrest, Avocado, WinkBed, Leesa and Tuft & Needle. We don’t currently earn commission from any of them; those pages link plainly to the brand’s own site, and we never imply otherwise. No brand, partner or not, can pay for a placement, an award or a score.

We also flag conflicts in our sources. Several of the largest mattress brands are owned by the same parent companies, and some review websites are owned by mattress companies. Where that affects a recommendation, we say so on the page. Our Canadian guide, for instance, notes that the owner of Douglas also runs mattress-review websites — and explains why we rank it first anyway, on the strength of independent testing and owner data.

Mattress data goes stale fast. Brands reformulate models without changing the name — RTINGS found two different coil designs sold as the same Saatva Classic — prices change weekly, and owner ratings drift. Every page shows the date its data was last checked; the current data set was researched in October 2026. When we update a model’s inputs, every score that depends on it updates automatically, because they all come from one data file.

How we verify prices. Prices come from the brand’s own product page (or, for showroom brands, the named retailer’s listing), checked on a stated date and recorded with that date. We record the typical selling price as well as the list price, because most online brands discount permanently, and value is scored on the typical price. When a price can’t be verified on a live page, we say so rather than carry an old figure forward.

We re-check prices, trials, warranties and owner ratings whenever we revise a guide, and we revisit sub-scores when a tester publishes a new review or a brand changes a model’s construction. When we get something wrong, we fix it and note what changed. If you spot an error — a price that has moved, a link that has died, a number that doesn’t match its source — please tell us via the about page.

Finally, a reminder that even a perfect score can’t account for time. Sleep Foundation suggests most mattresses last six to eight years, and that waking stiff or sleeping better elsewhere are signs it is time to replace yours. Durability is in the formula for that reason, but your own body is the best signal of when a mattress has had its day.