Why the legal limit is 4 parts per trillion

A few days ago I wrote about what "not detected" means on a laboratory report, and I ended with an admission. I had quoted EPA's PFAS drinking water standard, noted that the health goal is zero while the enforceable limit is 4.0 parts per trillion, and then said plainly that I had not opened the rulemaking record behind the specific figure of 4.0.

So I went and opened it. It is 226 pages of the Federal Register, and the answer is in there, stated more directly than I expected. It is worth the trip, because the reasoning turns out to be the cleanest real-world illustration I have found of something that is otherwise hard to convey: sometimes the number in the law is not a fact about the world. It is a fact about our instruments.

The sentence that explains the whole rule

EPA sets two numbers for a contaminant. The maximum contaminant level goal is the health target, set where no known adverse effects occur with a margin of safety. The maximum contaminant level is the enforceable standard, which the Safe Drinking Water Act requires to be as close to the goal as is feasible.

For PFOA and PFOS the goal is zero, because EPA concluded both are likely human carcinogens. You cannot enforce zero. So the question becomes what "feasible" means, and most people assume it means treatment: how clean can we actually get the water. That is part of it. But the rule identifies a second constraint, and then says which one binds:

"In practice this means that where the MCLG is zero, the EPA typically sets MCLs at the PQLs when treatment is otherwise feasible, based on cost and treatment availability, because the PQL is the limiting factor."

The PQL is the practical quantitation level. The rule defines it as "the lowest levels at which contaminants can be reliably quantified within specific limits of precision and accuracy during routine laboratory operating conditions using the approved methods."

Read the last clause of the first quotation again. The limiting factor is not the filter. It is the instrument. When the health goal is zero, the enforceable limit lands wherever measurement gives out, because a standard you cannot measure is a standard you cannot enforce. EPA says this about the MCL directly: in setting it, the agency "also identifies the level at which it is technologically feasible to measure the contaminant in the public water system."

So 4.0 parts per trillion is not a biological threshold. Nothing happens to a human being at 4.0 that does not happen at 3.5. It is the concentration below which the American laboratory system cannot reliably tell you what is in the water.

Whose laboratory, though

This is where it gets genuinely sharp, and where my own bias as someone who runs instruments had to be corrected.

My instinct on reading "measurement gives out at 4 ppt" was to object. Good mass spectrometry can see below that. EPA agrees, and says so in the same document: "measurements lower than the PQL are achievable by individual laboratories." Elsewhere it acknowledges outright that "lower quantitation levels may be achievable for some laboratories."

So the limit is not set by the best instrument. It is set by a quorum. EPA quotes the criterion it inherited from its monitoring programme, which defines a minimum reporting level as:

"a minimum quantitation level that, with 95 percent confidence, can be achieved by capable lab analysts at 75 percent or more of the laboratories using a specified analytical method"

Seventy-five percent of laboratories. That is the number hiding inside the number. The enforceable national limit for a likely human carcinogen is pinned to the performance of the American laboratory at roughly the 25th percentile, because a nationwide standard has to be one that water systems anywhere can actually get tested.

EPA is explicit about why it will not go lower, and the reasoning is administrative rather than chemical: a lower PQL "could potentially limit the number of laboratories available to support analytical monitoring." Tighten the number and you shrink the pool of labs qualified to check it. At some point the standard becomes unenforceable not because it is wrong but because too few places can run the test.

The methods are named: EPA Methods 533 and 537.1, both liquid chromatography with tandem mass spectrometry. The PQLs come from multi-laboratory validation studies conducted between 2017 and 2019, and EPA notes the calculations "account for differences in the capability of laboratories across the country."

That phrase is doing a lot. Differences in capability across the country are, in the end, what set the legal limit.

The proof by contrast

If the argument stopped there it would be suggestive but not conclusive. The same rule provides the control experiment, and this is the part that convinced me.

The 2024 rule regulated six PFAS, not two. For three of them, PFHxS, PFNA and HFPO-DA, EPA set the enforceable limit at 10 ng/L, which is exactly the same as their health goals. The stated reason:

"In addition, there are analytical methods available for these contaminants and the practical quantitation level (PQL) for each contaminant is below the level established by the MCLG."

The measurement floor for those three sat underneath their health targets, so measurement stopped being the constraint, and the enforceable number went straight to the health number. Same rule, same agency, same day, same laboratories. Where instruments could reach past the health goal, health won. Where they could not, the instruments set the limit.

That is as close to a controlled comparison as regulatory documents ever give you, and it settles what the 4.0 is made of.

The trailing zero is a claim

A small detail that I found delightful, and that will only really land if you have spent time arguing about significant figures.

The limit is written 4.0, not 4. That is deliberate and the rule explains it. EPA guidance, it says, requires that "all MCLs should be expressed in the number of significant digits permitted by the precision and accuracy of the specified analytical procedure(s)." The agency determined two significant digits were appropriate "considering existing analytical feasibility and methods," and quotes the methods themselves instructing that final concentrations be "rounded to an appropriate number of significant digits (one digit of uncertainty), typically two, and not more than three significant digits."

So the trailing zero in a federal regulation is not typography. It is an assertion about instrumental precision, encoded in the law. The regulation is reporting its own uncertainty in the notation, exactly as a measurement should.

What happens underneath the floor

Now the consequence, which is where this stops being a philosophical point about metrology and starts affecting what a water system is legally required to do.

Compliance is judged on a running annual average. The rule instructs that if a sample comes back below the PQL, "zero is used for that analyte solely to calculate the running annual average."

Consider what that means. A laboratory measures 3 ppt of PFOA. It is a real number, produced by a real instrument, describing a real molecule in real water. For the purpose of deciding whether the system complies with the law, that 3 becomes a 0.

I want to be fair to why this is done, because it is not a loophole. Below the quantitation limit the numbers are not precise enough to build an enforcement action on, and a regulator that penalised a utility using numbers it had already declared unreliable would lose in court and deserve to. The rule is being consistent: it declines to enforce on measurements it has declared unenforceable.

But the same rule refuses to let that zero spread. In the monitoring provisions it states that a system must use all results a laboratory reports, "whether above or below the practical quantitation levels," and then, unusually emphatically for a regulation: "zero must not be used in place of reported values."

The same measurement is therefore zero for one purpose and emphatically not zero for another. For compliance arithmetic it vanishes. For deciding how often you must keep testing, it counts. And the trigger levels that determine monitoring frequency sit at 2.0 ng/L for PFOA and PFOS, which is half the PQL. The rule triggers action on concentrations it will not use to find you in violation.

That is not a contradiction. It is a system that knows the difference between evidence good enough to act on and evidence good enough to punish with. Most disputes I see about detection limits, in environmental testing and in supplement certificates of analysis alike, come from collapsing those two standards into one.

EPA also rejected the argument that below-PQL results should be kept out of public reporting, saying it disagreed with commenters concerned about listing them as detected in consumer reports, because doing so "will allow customers to understand that the contaminant was detected in the water supply."

Where the rule stands in 2026

The regulatory picture has moved since 2024, and any honest reading has to say so.

EPA announced in May 2025 that it would keep the national drinking water limits for PFOA and PFOS, with Administrator Lee Zeldin saying the agency was "on a path to uphold the agency's nationwide standards to protect Americans from PFOA and PFOS in their water," while intending to rescind and reconsider the regulations for the other PFAS. On 18 May 2026 the corresponding proposals appeared. One would let systems request two additional years to comply with the PFOA and PFOS limits, moving the deadline from April 2029 to April 2031. The other would rescind the regulatory determinations and regulations for PFHxS, PFNA, HFPO-DA and the Hazard Index. The comment period on the rescission closed on 20 July 2026, and as of this writing both remain proposals rather than final rules.

Two things follow. The 4.0 ppt figure for PFOA and PFOS is untouched and now has a longer runway, so everything above about how it was derived remains live law. And my contrast case is a proposal away from disappearing. If the rescission is finalised, the three contaminants whose limits were set at their health goals will no longer be regulated at all. The reasoning EPA published about them stays on the record as reasoning, which is what I am using it for, but it will no longer describe an enforceable standard.

The honest limits on this

What I am not claiming. I am not saying 4 parts per trillion is unsafe, or safe. That is a toxicology question and this is a metrology argument. The health goal of zero reflects a carcinogenicity judgment I have not evaluated and am not qualified to second-guess.

I am not saying EPA did anything improper. The opposite: the agency wrote down its reasoning in the Federal Register in plain language, which is exactly why this piece is possible. Nothing here is a leak or an inference. It is quotation.

I am also not saying treatment cost played no role. The rule considers treatment technology and cost at length, and I have deliberately not summarised that analysis because I read the analytical-methods and MCL-derivation sections closely and did not work through the cost-benefit chapters with the same care. My claim is narrower than "cost did not matter." It is that EPA itself identifies the PQL as the limiting factor when the goal is zero, and says so in that word.

One boundary on the significant-figures point: I am reading EPA's stated rationale for expressing the limit as 4.0. I am not making an independent claim about the achievable precision of Methods 533 and 537.1, which would require the method validation data rather than the preamble.

The signal

There is a habit of mind worth taking from this, and it generalises well past drinking water.

When you meet a threshold in a regulation, a specification, or a certificate of analysis, ask what kind of number it is. Some thresholds are statements about the world: this is where harm begins, this is where the material fails. Others are statements about our ability to see: this is where our instruments stop being trustworthy. They look identical on the page. Both are just a number with units.

The PFAS rule is valuable because it puts both kinds side by side and labels them. Zero is the world. Four point zero is the instrument. The distance between them is not a compromise anybody negotiated, it is the width of our current ignorance, and it will narrow as methods improve rather than as chemistry changes.

Which leads somewhere slightly uncomfortable. If the enforceable limit tracks what laboratories can reliably do, then better instruments make the law stricter without anyone amending it. The standard for a likely human carcinogen is, in part, a moving readout of the state of analytical chemistry. That is a strange way to write a law. Given that you cannot enforce a limit you cannot measure, it is also difficult to see what the alternative would have been.