Why You Cannot Drill Out a Broken Tap

This page contains affiliate links. If you buy through one we may earn a commission, at no cost to you.

Not with any bit sold for the job, and the obstacle is a single number you can read off our own listings. A hand tap is hardened to 62 to 65 on the Rockwell C scale. The left hand drill bits in the extractor sets people reach for publish 63 on that same scale. A cutting tool needs to be meaningfully harder than the thing it cuts, and at 63 against 63 the margin is zero.

So the question splits in two. Removing it is an abrasive problem rather than a cutting one, or a chemical one where hardness never appears at all. And the reason it broke is unrelated to any of that: it is a cube law, and at small sizes the force that snaps a tap is under a fifth of a kilogram at each handle.

The hardness ladder, from figures our own records publish

Rockwell C numbers are convenient but they are not proportional, so the comparison has to be done in Vickers, where the scale is linear in indentation hardness. The standard conversion puts 62 HRC at 746 HV, 63 at 772, 65 at 832 and 67 at 900. Mild steel, the stuff most people are tapping into, is around 130 HV.

Our page on knife steel and hardness explains the scale correctly and gives the anchor worth holding on to: kitchen knives run 50 to 62 HRC, and the hardest knife steel in our catalogue is listed at 57 to 58. A broken tap is harder than every knife on this site, which is the first clue that the usual toolkit is out of its depth.

Now the tools, in the order people try them. A 15 piece stripped screw extractor set with left hand bits is the standard answer, and its record is unusually forthcoming about the one figure that settles this: seven "6542 HSS left-hand drill bits" with "hardness up to 63 HRC". 6542 is the designation for an ordinary molybdenum high speed steel, and 63 HRC is 772 HV. The tap is 772 HV. The hardness ratio is 1.00.

A ratio of 1.00 does not mean it cuts slowly. It means the edge yields at the same load as the material it is pressed into, so the bit rounds over, polishes the tap, work hardens the surface it failed to cut, and then snaps off inside the hole on top of the tap, which is a strictly worse problem than the one you started with. The working minimum for a cutting tool is around 1.5 times the workpiece in Vickers, and the handbook version of the same statement is that anything above 45 HRC wants carbide or ceramic.

Where our own guide gets the test wrong

We have a page on this and its advice is right for screws and wrong for taps, in a way that is worth quoting exactly because the error is in the kind of test it proposes. Our guide to removing a stripped or broken screw says: "Standard extractor bits handle carbon steel, cast iron, copper, and aluminum. Very hard alloys and 440-series stainless resist them, so identify the fastener before you commit."

Identify the fastener is good advice. Identifying it by alloy family is the wrong test, and a tap is the case that proves it. A hand tap is carbon steel or high speed steel by composition, which puts it squarely on the handled list. What makes it uncuttable is not its composition but its heat treatment: it has been quenched and tempered to 62 to 65 HRC because a thread cutting edge has to hold its geometry. Hardened 440C stainless, the material that page names as the hard case, is around 620 HV. The tap is 772, a quarter harder again than the example given as the limit.

So the test that works is hardness, not alloy. The same page's advice to drill counterclockwise and run slow is correct and useful, and the bits it proposes are the right bits for a snapped mild steel screw, where the ratio is 772 over 130 and nothing about this page applies.

The obvious upgrade, which also fails

If plain high speed steel will not do it, cobalt is the next shelf up. A 19 piece M35 cobalt stepped point drill bit set publishes an M35 grade, which carries about 5 percent cobalt and reaches 65 to 67 HRC, so call it 870 HV. Against the tap that is a ratio of 1.13. It is a real improvement and it is still well short of 1.5. In practice you get a little further into the job before the same failure.

That record also contains a defect that matters more here than it would anywhere else, and to its credit it reports the defect itself. Its own drawbacks note that the "Listing mixes M35 and M2 grade references". Those are not neighbouring grades for this purpose. M2 lands at roughly 63 HRC, which is a ratio of 1.00 and hopeless, while M35 lands at 65 to 67, which is 1.13 and merely insufficient. The listing cannot tell you which hardness turned up in the box, and on this job that ambiguity spans the whole difference between failing immediately and failing shortly afterwards.

What actually gets through it

Three materials clear the line, and each one has a different reason it may still not help you.

Tungsten carbide is the first. An 11 piece carbide masonry drill bit set publishes YG6X carbide, which runs around 1,600 HV for a ratio of 2.07, comfortably over the threshold. The problem is the shape rather than the material. These are brazed spear tips ground with a blunt, negative rake to crush brittle mineral, not to shear metal, and the record is honest about where that leaves them: "Not suitable for hard tile" and "Needs water cooling during use". Carbide is also brittle, and a broken tap sitting in a hole is a geometry that side loads whatever you put against it. The right carbide tool for this is a solid burr in a die grinder, and that is a shape our catalogue does not carry. We have the correct material in the wrong form.

Diamond is the second, and it is the one genuinely interesting exclusion on the chart. A 24 piece diamond drill bit set for glass carries diamond grit at something like 9,000 HV, a ratio of 11.7, which makes it eleven times harder than its problem and completely useless against it. Carbon dissolves into hot iron. Above roughly 700 degrees Celsius, which dry grinding reaches in the contact patch without trying, the diamond grit simply diffuses into the steel and disappears. The wear is chemical, not mechanical, and no amount of hardness touches it. Notice what that listing does and does not claim: sea glass, shells, gemstones, ceramic and stone, with no mention of steel anywhere, and a drawback that says "Hard material grinds slowly". That omission is correct and it deserves the credit. The hardest abrasive on this site is the one material on the chart that cannot be used.

Which leaves conventional abrasive, and it works. Aluminium oxide sits around 2,050 HV for a ratio of 2.66, it is bonded into a wheel rather than brazed to a tip so a fractured grain exposes a fresh one, and a 4-1/2 inch 9 amp angle grinder spinning a disc to 11,800 RPM will grind a hardened tap away without complaint. Its own record is appropriately blunt that it is "a genuinely dangerous tool that demands respect and safe technique", and our guide to using an angle grinder safely covers the handling.

The catch is geometric and it is absolute. A grinder can only reach a tap that stands proud of the surface. If the break is flush or below flush, which is the common case because the tap usually snaps where the torque is highest and that is at the workpiece face, then the one tool whose hardness is sufficient is the one tool that physically cannot get to the work. Everything that fits down the hole is too soft, and everything hard enough is too big.

The method where hardness does not appear at all

So stop fighting on that axis. If the tap is steel and the part it is stuck in is aluminium, the long standing shop answer is alum: potassium aluminium sulfate, the same compound sold for pickling, dissolved to saturation in water held near boiling. The solution is acidic and attacks iron and steel while leaving aluminium largely alone, because aluminium is protected by its own oxide and because the solution is already saturated with aluminium ions. The tap turns to sludge over hours to days and you rinse it out.

That approach is worth understanding as a category rather than a trick. Hardness is a mechanical property and it has no bearing on a chemical reaction rate, so by switching the mechanism you have deleted the entire obstacle rather than overpowering it. It also needs no tool at all, which is why it survives in machine shops that own every tool.

Its limits are real and worth stating plainly. It does nothing useful on a stainless tap, whose chromium oxide layer resists it. It will attack a steel or cast iron workpiece as readily as the tap, so it is only viable when the part is aluminium or brass. It is slow and the rate is unpredictable. And it wants a hot solution and ventilation, so it is a bench job rather than a kitchen one.

Why it broke, which is a cube law

None of the above explains the actual event, and the explanation is worth having because it is the only part of this you can prevent. A tap fails in torsion, and torsional capacity depends on the polar section modulus, which is pi times the cube of the core diameter divided by sixteen. The core at the flutes is roughly 0.57 of the nominal size on a four flute hand tap, so the torque a tap survives scales with the cube of its diameter.

Taking 1,200 megapascals as the torsional strength of hardened high speed steel, that gives 0.175 newton metres for a 1/16 inch tap, 1.16 for an M3, 2.87 for an M4, 5.45 for an M5, 9.26 for an M6, and 302 for a 3/4 inch tap.

Look at the span. A 4 piece adjustable tap wrench set publishes a range of "1/16 inch up to 3/4 inch", which is a size ratio of 12 and therefore a torque ratio of 12 cubed, exactly 1,728. One set of four handles covers a range over which the number you must not exceed changes by a factor of seventeen hundred, and the set publishes no torque figure anywhere, for any of the four. Nor does any other tap or tap wrench listing we carry.

Then convert the small end into something a hand can feel. A T bar wrench applies torque as two equal forces at the handle radius, so on a 100 mm span the 0.175 newton metres that breaks a 1/16 inch tap is 1.75 newtons at each hand. Under a fifth of a kilogram. Nobody can meter that, nobody can sense the approach of it, and the dual handle design that the record correctly praises for clamping "taps firmly" is the thing that makes the limit invisible, because a symmetric grip feels controlled right up to the moment the tap is in two pieces.

At the other end an M10 tap wants about 44 newton metres, which is a genuine heave, and that is why large taps rarely break and small ones break constantly. The ceiling falls as the cube of the size. The handle gets shorter roughly in proportion to the first power. And the strength of a hand does not change at all.

What pushes the torque over the edge

One more piece of arithmetic, because it names the moment of failure. A tap has to store its own chips. Cutting an M6 by 1 thread through 12 mm of plate removes an annulus of pi over four times 36 minus 25 square millimetres, which is 8.64 square millimetres of section, or 103.7 cubic millimetres of metal over that depth. A four flute M6 tap offers perhaps 6 square millimetres of flute cross section in total, which over the same 12 mm is around 72 cubic millimetres of space. Swarf is loose and bulks up at least twofold against solid metal.

So the flutes are something like three times over capacity before the thread is finished. Once they pack, the chips are no longer being carried out, they are being compressed between the tap and the thread it just cut, and the torque climbs steeply for no change in cutting. That is the spike that finds the cube law. Backing off every half turn is not superstition, it is the only way to empty a container that was never large enough, and the flute cross section is an estimate because no tap listing publishes one.

The prevention is already in the catalogue, sold for something else

Put the two sets of numbers side by side and one tool in our catalogue turns out to be the right instrument for a job it is not advertised for. A 90 piece master torque screwdriver set publishes a 1/4 inch driver with a "10 to 70 in-lb range in 0.5 in-lb steps" and an audible click at the preset. Convert that: 10 in-lb is 1.13 newton metres and 70 in-lb is 7.91, in steps of 0.057.

Against the ladder above, the mapping is exact. At its full scale of 7.91 newton metres it is below the 9.26 that breaks an M6, so on an M6 tap the tool is physically incapable of breaking it no matter how you lean on it. Set to 20 in-lb, which is 2.26 newton metres, it sits at 79 percent of an M4's limit. Set to 40 in-lb it protects an M5. Below M3 it cannot help at all, because its minimum setting of 1.13 newton metres is already 97 percent of the 1.16 that destroys an M3, and the whole 1/16 inch end of the range is two orders of magnitude below anything it can read.

The listing's own stated uses are "gunsmithing, scope mounting, and bicycle repair", and tapping is not among them. Its drawbacks say it is "Overkill for casual assembly tasks", which is true of assembly and the opposite of true here.

The honest limit

The weak number on this page is the torsional strength of the tap. 1,200 megapascals is a representative figure for hardened high speed steel in torsion and no tap listing publishes a strength, a hardness, a core diameter or a torque rating, so every value in the lower panel of the figure rests on it. Scale it up or down and all six bars move together, which is why the ratios between them, and the 1,728 across one wrench set, are the trustworthy part.

The hardness side is firmer, because the 63 HRC that makes the whole argument is printed in our own extractor listing rather than assumed, and a ratio of 1.00 does not need precision to be conclusive.

And the technique that beats every purchase on this page needs nothing: tap the hole with the tap held in a drill chuck or a tapping block so it starts square, then use a tap one size up in two stages on anything below M4, and back off half a turn every half turn. Alignment and chip clearance are free, they address the two causes above directly, and they are the only part of this problem that is still open once the tap is already broken off below the surface.