Technical guide
Repairing an existing screed: bringing an old floor up to standard
Most existing floors are not failed floors. They were laid to the standards and by the methods of their day, and what has moved since is the specification of everything laid on top of them. This guide covers the defects that actually occur, how each is identified, and what brings a floor back within the 3 mm tolerance modern coverings demand.
By Mihail Cheptene · 10 August 2026
What changed is not the floor
Almost without exception, a building with a reinforced concrete base has a screed: a sacrificial layer laid because the slab comes out of the formwork as structure rather than as a surface, cast to tolerances that were never surface tolerances and short of the levels, falls and service runs the finish is set out from. That layer is now under most of the flooring in the country, and in the great majority of cases it is doing its job.
These floors were poured for quarry tile, lino and thick-bed ceramics, and judged by a standard that has barely moved since. What moved is everything above it. Today's coverings are porcelain slabs in three-metre sheets, LVT, glued engineered board: thin, rigid, and rectified to tolerances that leave nothing to absorb a ridge. They follow the substrate exactly and report what they find. So the floor did not fall below its own standard — it fell below the standard of what now goes on top of it, and the question is not whether it can be saved but which defect is present and which repair answers it.
Repair against removal
Breaking out 60 or 70 mm of screed means skips, dust through the property, a fresh pour, and a drying period that starts from zero: in a conventional sand:cement screed, a week under polythene, then three weeks before ceramic tiling — and where a moisture-sensitive covering follows, a day per millimetre for the first 50 mm and slower beyond it. Rapid-hardening cements shorten that at a cost per bag. A repaired floor is measured against the same tolerances and either meets them or does not, but it waits only for what was added to it: the floor itself has been in equilibrium with the building for years.
Identification is the work
The quality of a repaired screed depends on one thing: whether the problem was identified properly. Every material solves a specific failure and does nothing for the others, so choosing the material is not the skill — naming the failure is. These floors were mixed and levelled by hand, without plasticisers, fibre or a power float, which is why the variation is local rather than general: the floor is sound over most of its area and wrong in defined places.
Four things occur. A crack, where the floor shrank and had nowhere to move. Hollowness, where the screed never gripped the slab, or where nothing was ever meant to grip in an unbonded or floating construction. A weak, dusty surface, where too much water left a soft skin over a solid core. And a previous repair that failed — most often a levelling compound poured over cracks that were never dealt with. The first two are read off the surface, a crack by eye and hollowness by tapping; what the other two are worth is settled by opening the floor. That last one misleads most reliably: the compound is thin and bonded tight, so a crack underneath comes straight through it. Lifted, it shows one of two things — a sound screed with a defect running through it, or a weak layer that came away on the underside of the compound and has to come off.
One of the four decides the method rather than describing it: a weak surface over a solid core is three to five millimetres taken off, while a floor weak throughout is a floor removed — and the difference is found by cutting into it, not by scratching the top. Age settles nothing either. A floor laid three years ago can be in worse condition than one from the seventies that followed the method properly.
Cracking through the full area with the sides sitting at different levels. Local repair answers local defects; this is the floor itself.
Cracks: stitched across, not filled along
The crack is opened along its line with a diamond blade, and transverse slots are cut across it — around 80 mm long at roughly 250 mm centres, to a depth of up to two thirds of the screed thickness. Steel connectors are laid into the slots across the line and the whole is filled with low-viscosity resin, sand broadcast in while it is wet to give the next layer something to key into.
The steel is what holds the two sides together; the resin bonds it and fills the void. A cementitious filler is the wrong material here, because shrinkage is the mechanism being repaired and a shrinking filler repeats the failure. Where a crack has landed close to where a joint belongs, opening it out and treating it as a joint is better than stitching it and cutting another 300 mm away — the floor has already shown where it wants to move.
Steel laid across the crack in transverse slots before the resin goes in. The bars hold the two sides; the crack itself is not what is being filled.
The same line filled with resin, still wet. A shrinkage crack running clear across the middle of the room.
The slots seen along the line: cut across the crack rather than along it, at roughly 250 mm centres, each one taken to about two thirds of the screed depth before the bar goes in.
Removed areas: square edges, matched material
Debonded, broken or weak material is cut back with a vertical edge and removed entirely. The perimeter is sawn rather than chased out, because a feathered edge fails at its own thin end and reads as a fresh defect a year later. Everything loose comes out, the surface is vacuumed, and a primer or bonding slurry goes on before the mortar. Rapid-hardening screed cements rebuild to the full depth of the existing screed and take tiling within about a day.
Steel is not part of a straightforward patch: repair mortars are formulated close to the parent material in strength and stiffness precisely so the interface is not stressed. Steel enters where the patch crosses a crack — a continuation of the stitching, not a property of the patch — and on large areas or edges carrying dynamic load.
Weak surfaces and failed compound
A soft, dusty top over a solid core is taken off mechanically, ground or scabbled back to sound material. Consolidating primers have a place where dust is the problem rather than strength; they stabilise a surface, they do not carry load. Resin impregnation can rebuild a moderately weak layer, but it is a specification in its own right, priced and applied as one and dependent on the layer being thin and absorbent — it is not a primer coat asked to do more than it was sold for.
Levelling compound coming away with the top of the screed still attached to its underside. The primer held; what it was bonded to was the weak layer, and that is what has to come off.
Broken levelling compound comes off, and then the floor is diagnosed as if it had never been there — because the defect that broke it is still present. Cracks are stitched, hollow material is removed, and only then does a new levelling layer come into question, if it comes into question at all.
The same layer broken out where it had cracked. Underneath, the cause is visible: an unstitched crack, now marked out for transverse slots.
Bringing it to tolerance, if it needs it
BS 8204-1 classifies surface regularity by the maximum departure from a 2 m straightedge under its own weight: 3 mm, 5 mm and 10 mm for SR1, SR2 and SR3. Tiling works to SR1, and SR1 is the tightest class the standard has — large-format porcelain is specified against it because there is nothing above it to specify. Where the screed was flat to begin with and the repairs are flush, the work ends there: pouring compound across a floor already in tolerance adds cost, weight and height for nothing.
Where the plane is genuinely out, one thin pour across the whole room beats levelling each area individually, because a tile bedded across two different surfaces behaves differently on each side. Its limits decide when this stops being the answer: it follows what is underneath, so over a hollow it gives a smooth hollow and over an unrepaired crack it breaks along the crack. Beyond roughly 20 to 25 mm it has stopped being a levelling question and become a relaying one.
Residual movement: what uncoupling covers
An uncoupling membrane breaks the bond between tile and substrate so small residual movement is not transmitted into the tiling. Over a stitched crack it is a reasonable final layer. It absorbs a modest amount of horizontal movement and no vertical movement, bridges no voids, adds no strength, and does not remove the perimeter and movement joints the tiling standard requires. Its position is fixed: over the levelling layer, not under it.
Common questions, straight answers
The floor sounds hollow in places. Does it all have to come up?
Rarely. On an unbonded or floating screed, hollowness is the construction working as designed. Even on a bonded floor it is not in itself a defect: what matters is the extent, the position and whether the separated material is sound, flat and thick enough to carry what goes on it.
There is already levelling compound down and it is breaking up. Is the screed finished?
Usually the opposite. Compound breaks up because something underneath moved or was never strong enough to bond to, and the compound had no capacity to absorb it. Lifted, it shows which: a sound screed with an unstitched crack through it, or a weak top layer that came away on the underside of the compound. The first is stitched, the second is ground off. Neither is a reason to take the floor out.
How long before tiles can go down over a repair?
Governed by the repair materials, not by the screed. Resin stitching and rapid-hardening cements generally allow tiling the following day; a levelling compound follows its own data sheet for the depth poured. The three-week rule belongs to a new pour in conventional sand:cement.
Does every repaired floor need a levelling compound over it?
No. If the repaired floor meets 3 mm under a 2 m straightedge it is ready as it is. A compound corrects plane; it adds nothing else.
Does an uncoupling membrane mean the cracks can be left?
No. Cracks are stitched first; the membrane then covers the residual movement in a floor that is otherwise sound. It manages what is left after the repair, it does not hold two pieces of screed together.
Referenced to BS 8204-1:2003+A1:2009, BS 5385-1:2018 and BS 5385-3, with manufacturer data for repair resins, rapid-hardening screed cements and uncoupling systems. Standards set the acceptance criteria; product data sheets set working times and permitted combinations.