Membrane treatment is best understood as one continuum rather than four products. Microfiltration, ultrafiltration, nanofiltration, and reverse osmosis differ in what they remove, what pressure they demand, and what they cost to run, but they share a logic: put a barrier in the water's path and let physics do the sorting. Reading the spectrum correctly is the foundation of sensible selection, municipal or residential.
Four classes, one axis
Moving across the spectrum, the barrier gets tighter, the targets get smaller, and the energy bill gets larger.
| Class | Character | Representative targets | Relative pressure |
|---|---|---|---|
| Microfiltration (MF) | Porous, on the order of a tenth of a micron | Sediment, algae, most bacteria, protozoan cysts | Low |
| Ultrafiltration (UF) | Porous, roughly an order of magnitude tighter | Colloids, most viruses, larger organics | Low to moderate |
| Nanofiltration (NF) | Transitional, partially rejecting | Divalent ions such as hardness, color, many larger organics | Moderate |
| Reverse osmosis (RO) | Dense, diffusion-controlled | Dissolved salts and most solutes broadly | Highest |
The MF and UF classes are genuinely porous filters, and their removal story is fundamentally about size. NF and RO are better understood as dense barriers where water dissolves into the membrane material and diffuses through; rejection there depends on solute charge and chemistry as well as size, which is why NF can pass much of the sodium while rejecting most of the hardness, a behavior no simple sieve would produce.
Concepts that carry across the spectrum
A handful of terms do most of the explanatory work in membrane conversations. Flux is the rate of water through a unit of membrane area. Recovery is the fraction of feed water that becomes product rather than concentrate. Rejection is the fraction of a given solute held back. Every membrane system trades these against each other and against fouling, the accumulation of material on or in the membrane that raises pressure requirements and shortens cleaning intervals. Pretreatment exists to manage fouling before it happens, and scaling, the precipitation of sparingly soluble minerals in the concentrated boundary layer, is the specific fouling mode that punishes high-recovery operation on hard water.
Two practical corollaries follow. First, membranes are barriers, not destruction technologies: everything rejected ends up somewhere, and the concentrate or backwash stream is part of the design rather than an afterthought. Second, integrity matters as much as tightness: a compromised fiber or a failed o-ring converts an excellent barrier into a plumbing fitting, which is why municipal membrane plants run routine integrity testing and why residential membranes deserve periodic performance checks rather than indefinite trust.
Where each class earns its keep
In municipal practice, MF and UF have become standard tools for particulate and pathogen barriers, frequently serving as pretreatment protecting downstream RO in reuse and desalination trains. NF finds work where the problem is hardness, color, or organics rather than total dissolved solids. RO anchors seawater and brackish desalination, potable reuse, and high-purity industrial supply. Concentrate management, meaning what to do with the stream that carries everything the membrane rejected, is a siting and permitting question that deserves attention as early as process selection itself.
Residentially, the spectrum compresses to two familiar offerings: UF-based point-of-use units where the goal is particulate and microbial polishing while keeping minerals, and point-of-use RO where the goal is broad reduction of dissolved solids, a subject we treat in detail in our RO explainer. The same selection logic applies at both scales: name the target, then choose the loosest membrane that credibly addresses it, because every unnecessary step up the spectrum is paid for in pressure, reject water, and maintenance.
Membrane selection goes wrong in predictable ways: chasing a dissolved contaminant with a porous filter, or deploying RO against a problem UF would have solved at a fraction of the operating cost. The spectrum, read as a single axis of tightness versus cost, is the antidote to both mistakes.