Membrane Filtration in Water Treatment: Picking Between MF, UF, NF and RO

Four membrane classes get sold under the one heading, and only two of them do anything to dissolved salt. Microfiltration and ultrafiltration strain water through actual holes. Nanofiltration and reverse osmosis push it through a dense polymer film where separation happens by diffusion. Treat all four as rungs on a single ladder of pore sizes and you'll specify the wrong class, almost always in the direction that costs money.

EPA draws that line in its Membrane Filtration Guidance Manual, and the wording repays a close read. Microfiltration and ultrafiltration are each defined as a pressure-driven membrane filtration process, with a pore size attached: roughly 0.1 to 0.2 microns for MF, 0.01 to 0.05 microns for UF. Nanofiltration gets no pore size at all. EPA calls it a pressure-driven membrane separation process that employs the principles of reverse osmosis, applied for membrane softening or removing dissolved organics. The second half of EPA's RO definition is that same sentence, word for word. The manual still files all four under membrane filtration as an umbrella term, but on the question of what the process does, it put NF on the RO side of the fence.

That shows up in your permeate. A nanofiltration element can pull better than 97 percent of the sulfate and magnesium out of a feed and still hand you water carrying several hundred milligrams per liter of sodium chloride. If the binding line in your acceptance spec is hardness, that's a win at a fraction of the pumping cost. If it's conductivity, you bought the wrong membrane, and no amount of pressure will fix it.

Two of them strain, two of them diffuse

MF and UF are porous barriers. They're rated by pore size, they're integrity-testable, and they run cheap. DuPont's IntegraTec SFP-2860 module is 0.03 micron nominal PVDF fiber with 549 square feet of area, and the data sheet puts typical filtration transmembrane pressure at 0.4 to 1.5 bar, call it 6 to 22 psi, with a hard cap at 2.1 bar. That's the whole pressure envelope. It takes out colloids, suspended solids, bacteria and protozoan cysts. Sodium goes straight through, and raising the pressure doesn't change that, because there's a hole and the ion fits.

NF and RO have no hole to fit through. Water dissolves into the polyamide layer and diffuses across; most ions don't. Rejection becomes a function of ionic charge and size and the chemistry of the film itself, so a divalent ion like sulfate gets stopped far more completely than chloride at the very same membrane. That charge dependence has no analogue in MF or UF, and it's what makes NF behave like a selective RO rather than a coarse one.

That 0.03 micron figure is a nominal rating, by the way, and it lands inside EPA's ultrafiltration band. Nominal is doing work in that sentence. It describes roughly where the pore distribution centers, not a guaranteed largest pore, so it tells you what the module is built to do rather than what a regulator will credit it for.

What each class leaves behind

One caution on the virus row. A log-removal credit comes from a validated module plus an integrity-testing protocol, and that is the only place a regulator will accept one from. A nominal pore size rating has no standing in that conversation.

Pressure is what you pay for the last few points of rejection

Put a nanofiltration element and a brackish RO element side by side on their published test conditions and the trade gets blunt. DuPont tests the NF270-4040, a 4-inch element, at 70 psi on 2,000 ppm magnesium sulfate and publishes 2,500 gpd from 82 square feet at better than 97 percent rejection. NanoH2O (formerly LG) tests the 8-inch BW 400 R G2 at 225 psi on 2,000 ppm sodium chloride and publishes 11,500 gpd from 400 square feet at 99.8 percent stabilized, with a 99.65 percent floor. Different diameters, so the raw gpd figures aren't the comparison. The area-normalized flux is.

Work those back to flux. The NF element makes about 30 gallons per square foot per day at 70 psi. The brackish RO element makes about 29 at 225 psi. Call it the same water per square foot at roughly a third of the applied pressure. Per psi you put in, the NF membrane is about 3.4 times as permeable, and pump pressure is a bill you pay every hour the plant runs. That's the argument for nanofiltration and it's a strong one. Both fluxes are bench numbers on clean synthetic feed, mind you, and design flux on real water runs a fraction of either. The ratio is the point, not the absolutes.

The rejection figures, though, aren't comparable, and nobody should pretend otherwise. DuPont's 97 percent is measured on magnesium sulfate, a divalent salt that any NF membrane stops easily. NanoH2O's 99.8 percent is measured on sodium chloride, the hard case. Reading those two numbers straight across is the standard way to talk yourself into the wrong element, and it's the same trap as comparing flow numbers without reading the test conditions.

Two elements both called nanofiltration

NF270 and NF90 share a spec sheet layout and not much else. Same 82 square feet, same 70 psi test pressure, same 2,000 ppm magnesium sulfate at 15 percent recovery. The NF270 returns 2,500 gpd at better than 97.0 percent. The NF90 returns 2,000 gpd at better than 98.7 percent. So the NF270 gives you 25 percent more permeate for about 1.7 points of magnesium sulfate rejection.

On monovalent ions the gap runs far wider than that comparison implies, and DuPont says so in print. Its NF270 flat sheet data sheet describes the membrane as having excellent rejection of divalent ions and of organics above 400 amu, while featuring monovalent rejection of about 40 to 60 percent. Two elements, one product category, and sodium passage that differs by a factor you can taste. Writing "nanofiltration" on a bid line specifies almost nothing. The element number is the specification.

The failure mode: softening numbers bought against a TDS target

This one arrives as a membrane complaint and turns out to be a specification error. Somebody has a hardness problem, watches NF soften beautifully at a third of the pressure, writes NF into the design, and then the acceptance test gets run on conductivity.

Run the arithmetic before the purchase order, because it isn't close. Take a 1,000 mg/L feed that's mostly sodium chloride. At the 40 to 60 percent monovalent rejection DuPont publishes for NF270, permeate lands somewhere around 400 to 600 mg/L. The BW 400 R G2 at its 99.8 percent stabilized rejection puts the same feed near 2 mg/L, and about 3.5 at its published minimum. Those are answers to different questions, and a blending valve won't rescue the first one.

Both figures are bench numbers, and the benches aren't even the same bench. The RO side comes off an element run at 15 percent recovery. The NF monovalent range comes off a plate and frame coupon at 130 psi with no recovery stated at all. Put real elements in a real array at 70 or 80 percent recovery and the feed-brine average concentration climbs well above the feed, so both permeates land higher than the sheet says. The gap between them survives the trip. The absolute figures don't, so don't carry 2 mg/L into an acceptance test.

The tell is available long before commissioning: find the line in the acceptance spec that binds. Hardness, sulfate, color, TOC and THM precursors are nanofiltration territory. Conductivity, TDS, sodium, chloride, boiler feed quality and pretreatment ahead of deionization are RO. When both bind, NF followed by a small RO polish usually beats one oversized train, though that's two line items and the budget needs to hear it early. The same logic sorts the low-pressure end, where the mistake runs the other way and UF gets asked to do a dissolved-solids job it has no mechanism for. We've laid that pairing out separately in ultrafiltration versus reverse osmosis.

Reading your own acceptance spec

Start from the binding line and work backward to the class, rather than starting from a membrane and hoping it lands. Particles, turbidity or pathogens: UF does it at single-digit to low-twenties psi across the membrane and you're finished. Hardness, sulfate, color or TOC with a loose TDS target: nanofiltration gets most of the removal at a third of the pumping cost, and specifying RO there buys rejection nobody asked for. Conductivity or TDS binding: you're in brackish RO, and the number to shop on is the rejection floor rather than the headline. NanoH2O's own 8040 sheets show why. The BW 400 R G2 and the premium MaxRO R both print 99.8 percent stabilized at 11,500 gpd from 400 square feet on the same 225 psi test, and they separate on the floor, where the R G2 holds 99.65 percent against MaxRO R's 99.5. MaxRO R earns its premium somewhere else, on a 36 mil ultra-low differential pressure spacer that pays off on fouling-prone feed and in long vessels. The high-flow BW MOST buys 13,200 gpd at 125 psi, so it's more water for roughly half the applied pressure, and it drops to a 97.0 percent minimum to get there. Three different test pressures in one lineup, so compare them on your own design flux rather than on sheet flows. Where permeate quality binds on an 8040 brackish rack, the R G2 is the call, and FilmTec stays the shop default on installed base and contractor familiarity rather than on a rejection number.

Enviro Membranes stocks NanoH2O 8040 elements, the BW 400 R G2, BW 440 R G2 and SW 400 R, with same-day US shipping, so a rack that needs elements isn't idle waiting on a lead time. If you're weighing nanofiltration against brackish RO for a specific feed and a specific acceptance spec, send us the water analysis and the binding numbers and we'll tell you which class the job needs, even when that's a membrane we don't sell.