Why stockinette curls
Knit a rectangle of plain jersey and it will not lie flat. The top and bottom edges roll toward the back; the two side edges roll toward the front. Knit the same yarn as garter stitch and the rectangle lies flat as a board.
The cause is entirely structural, and it can be counted.
The loop is not symmetric
A knitted loop has three parts: two legs running down into the loop below, and a head running across the top where the next loop passes through. The legs sit on one face of the fabric and the head sits on the other.
That is the asymmetry, and it is unavoidable. Nothing about the operation of drawing one loop through another can be made front-to-back symmetric, because the yarn has to come from somewhere and go somewhere.
The consequence is that a stitch pulls differently on the two faces. Along the top and bottom of the fabric the heads are on one side, so that side is under more tension and the edge rolls toward it. Along the two sides the legs dominate, so the roll goes the other way.
Which way each edge goes
The directions are worth stating because they are consistent and slightly surprising.
In stockinette, the cast-on and cast-off edges — the ones running along a course — roll toward the purl side, which is the back of a jersey fabric. The selvedges — the ones running along a wale — roll toward the knit side, the front.
So the four edges of a jersey rectangle do not roll the same way. Two go back and two go front, and the corners are consequently doing something complicated. Anybody who has laid out a piece of jersey has seen it: the thing does not curl into a tube, it curls into a shape with an argument at each corner.
That is a genuinely structural prediction rather than an observation dressed up as one. The heads and the legs run in perpendicular directions, so the imbalance they produce acts along perpendicular axes and has opposite senses.
Where the asymmetry comes from
It is worth being precise about the geometry, because “the loop is asymmetric” is easy to say and easy to say vaguely.
A loop has two legs descending into the loop below and a head across the top through which the next loop passes. The legs are on one side of the fabric plane and the head is on the other — necessarily so, since the head has to be reachable by the next loop from the far side.
So each stitch has more yarn on one face than the other, at the top and bottom of the loop respectively. Along the fabric’s course direction the heads dominate one face; along the wale direction the legs dominate the other. Two imbalances, acting along perpendicular axes, with opposite senses.
That is why the four edges of a jersey rectangle do not behave alike, and it is a prediction rather than a description: the sense of the curl at each edge follows from which part of the loop dominates there.
What cancels it
Once the cause is stated the remedy is obvious: arrange the courses so consecutive ones face opposite ways, and the imbalances cancel.
Garter stitch does exactly that: knit one course, purl the next. Every course’s asymmetry is opposed by its neighbour’s, and the fabric lies flat.
Rib does the same thing across the fabric rather than along it: alternate knit and purl within a course. That cancels the sideways imbalance, which is why a rib does not curl at its selvedges, and it also gives rib its enormous crosswise extension because the knit and purl wales fold toward opposite faces.
Seed and moss stitches alternate in both directions at once and lie flattest of all.
Interlock is two rib fabrics knitted together and is symmetric by construction, which is why it does not curl at all and why it is used where stability matters.
Counting it
The claim can be reduced to a number, which is what the figure on this page does.
For each fabric, go down the courses and count how many times the facing changes from one course to the next. Stockinette gives zero — every course faces the same way, so every asymmetry adds. Garter over six courses gives five — every consecutive pair opposes.
That count is a structural property of the arrangement, computed from the pattern rather than from the fabric, and it predicts the mechanical outcome exactly: zero changes means curl, and a change at every course means flat.
It is worth being careful about what has been established. The count is exact and the mechanical consequence is an inference: nothing here computes a bending moment or predicts a radius of curl. What the count shows is that the two fabrics differ in exactly the structural respect the explanation invokes, and that the difference does not depend on yarn, gauge or size.
Why blocking only half works
Curl can be suppressed and it is difficult to remove, and the difference is instructive.
Blocking — wetting the fabric, pinning it flat and drying it — sets the yarn into its pinned shape, and for some fibres that is fairly permanent. Wool takes a set well because its scales and its disulphide bonds can be persuaded to reform; linen and cotton hold a wet-set shape moderately; synthetics hold a heat-set shape very well and a wet-set one hardly at all.
But blocking has not removed the asymmetry. It has added a competing stress that happens to oppose it, and the underlying imbalance is still there. Wash the piece again without pinning and a wool jersey will curl again, less than before.
The structural remedies do remove it. A garter or seed border, a ribbed band, a facing knitted in interlock — all of these change the arrangement so that there is nothing to cancel. That is why every knitting instruction that has ever been written begins a stockinette piece with a few courses of something else.
What the count does and does not establish
Being careful here matters, because this is the one essay on this site where a structural count is being connected to a mechanical outcome by argument rather than by computation.
What is computed: the number of times the facing changes from one course to the next. Zero for stockinette, one per course for garter. That is exact, it follows from the arrangement alone, and the figures on this page report it.
What is inferred: that zero changes means the asymmetries add along the edges and produce a bending moment, and that alternating changes means they cancel. That is an argument about where the stresses are, not a solved elasticity problem.
What is not attempted at all: the radius of the curl, the force required to flatten it, or how any of it varies with yarn stiffness and gauge.
So the honest claim is narrower than “this explains curl”. It is that the two fabrics differ in exactly the structural respect the usual explanation invokes, that the difference is exact rather than approximate, and that it does not depend on yarn, gauge or piece size — which is consistent with the observation that stockinette curls in every yarn anybody has tried.
That is a weaker claim than the ones this site makes about weaving, where integrity and float length are computed outright. Marking the difference is more useful than blurring it.
Where the woven analogy fails
It is tempting to look for the same effect in weaving, and it does not exist in the same form.
A woven cloth’s two faces can be very different — a satin or a 3/1 twill is warp on one side and weft on the other — and such cloths do have a preferred face and can cup slightly if the two systems have different shrinkage. But they do not roll at their edges the way jersey does, because the woven structure has no equivalent of the loop’s head-and-legs asymmetry. The threads are straight and symmetric about their own axes; only their arrangement differs between the faces.
What woven cloth has instead is fraying, which is the failure mode of an unsupported edge in a structure of separate threads. A knit does not fray and a weave does not curl, and both are consequences of the same difference in construction.
Why it does not go away with size
A property worth noticing, because it distinguishes this from the sort of edge effect that becomes negligible in a large piece.
Curl is an edge phenomenon, so a first guess is that a large piece of jersey would be mostly flat with a curled border. That is roughly what happens — but the border does not get relatively smaller in any useful sense, because its width is set by the yarn and the gauge rather than by the piece.
The reason is that the imbalance is per-stitch and acts over a few stitches from the edge. A stockinette square four inches across and one four feet across roll at their edges by about the same physical distance, so the large piece is proportionally flatter and the small one is a tube. Anybody who has knitted a swatch has met this: a tension square of jersey is nearly unmeasurable because the curl consumes most of it.
This is also why a border works. A few courses of garter or rib at the edge put the cancelling arrangement exactly where the imbalance acts, and a border two or three stitches wide is often enough — which would be surprising if the curl were a whole-fabric property and is not surprising at all if it is an edge one.
Curl put to work
A defect in one context is a mechanism in another, and knitted design uses the curl deliberately.
A stockinette edge left unfinished makes a rolled edge that needs no hem — used constantly in machine-knitted garments, where it saves an operation and looks intentional. The roll is consistent enough to be specified.
More substantially, the same asymmetry that curls an edge is what makes a rib pull in. A rib’s knit and purl wales fold toward opposite faces, so the fabric concertinas, and its relaxed width can be half its extended width. That is not a side-effect; it is the entire reason cuffs and waistbands are ribbed, and it is the loop’s front-to-back asymmetry acting sideways instead of along.
Heat setting, and the fibre that does not care
There is one way to defeat curl outright rather than cancelling it, and it belongs to the material rather than the structure.
Heat setting takes a thermoplastic fibre — nylon, polyester — above its glass transition while the fabric is held flat, and lets it cool there. The polymer chains rearrange into the held configuration and the fabric’s memory is rewritten: it is now flat by preference, and the loop asymmetry has nothing to push against.
That is why machine-knitted polyester jersey lies flat straight off the machine and cotton jersey does not. It is also why heat-set fabric cannot easily be reshaped afterwards, and why a garment made from it holds its finished dimensions well.
Natural fibres take a set imperfectly. Wool is the best of them, because its disulphide bonds can be broken and reformed by steam — which is why wool tailoring can be shaped by pressing and cotton cannot. Cotton and linen hold a wet-set shape only until they are wet again.
So the honest summary is that curl is structural, and whether it shows depends on whether the fibre will accept a competing instruction. That is one of the places on this site where structure hands over to material, and it is worth marking them when they occur.
What this account does not do
Three limits, and the second is the honest one.
It does not predict the radius. How tightly a given jersey curls depends on yarn bending stiffness, loop length, gauge and fibre, none of which is computed here.
It does not derive the mechanics. The connection between loop asymmetry and edge curl is an argument about where the stresses are, not a solved elasticity problem. The figures on this page count a structural property and do not compute a force, and calling the count a proof of the mechanics would be over-claiming.
It does not cover warp knits. Tricot and raschel fabrics have loops too and mostly do not curl, because their loop arrangement is different and their structure is far more constrained. Almost every generalisation about knits on this site is a statement about weft knits.
The same asymmetry, used deliberately
Two structures exploit the imbalance rather than cancelling it, and both are everywhere.
Rib alternates knit and purl within a course. The knit wales fold toward one face and the purl wales toward the other, so the fabric concertinas and its relaxed width is a fraction of its extended width. That is the loop asymmetry acting sideways, and it is the entire reason cuffs, collars and waistbands are ribbed — the fabric supplies its own elastic recovery without any elastic in it.
Purl — alternate knit and purl courses within a wale — does the same lengthways, giving a fabric that concertinas vertically. It is much less used because vertical stretch is rarely what a garment needs.
The pattern to notice is that the loop has one asymmetry and it produces several apparently unrelated effects — curl, rib’s pull-in, the way a knit fails — depending on which direction it is allowed to act in. That is a good sign that the structural account is the right one: a single cause with several consequences is more convincing than several causes fitted to several observations.
The rib, which is the same thing sideways
The clearest confirmation that the explanation is right is that the same asymmetry, allowed to act in the other direction, produces a completely different and equally familiar effect.
Rib alternates knit and purl within a course rather than between courses. So along the fabric’s width, consecutive wales face opposite ways — and instead of an edge curling, the fabric itself folds. The knit wales come forward and the purl wales go back, the cloth concertinas, and its relaxed width is a fraction of its extended width.
That is where a cuff’s grip comes from. There is no elastic in a plain rib; the recovery is the yarn’s bending elasticity acting through a structure that has folded itself, and it is enough to hold a sleeve at a wrist.
One asymmetry, three consequences depending on the direction it acts in: curl at an unsupported edge, flatness when it alternates along the wale, and pull-in when it alternates along the course. A single cause with several apparently unrelated effects is a good sign that the cause is the real one.
Where the ladder goes next
The other consequence of the loop’s structure is how a knit fails, where the topology rather than the asymmetry does the work.
The structure itself is the loop, and the general principle that arrangement dominates material is where this site starts.
What the pictures here cannot show. The figures on this page draw loops as schematic curves and count how the courses face. They do not draw a curled fabric, because curl is a three-dimensional consequence of a stress imbalance and these are flat diagrams of a structure.