A fillet weld is described by two different dimensions and they are not interchangeable. The drawing calls out a size, the designer sized the connection on a throat, and the gauge in your hand reads one of them and not the other. Confusing the two is the most common source of arguments about whether a fillet is big enough, and it is a distinction examinations keep returning to because it separates people who have measured welds from people who have read about them.
This guide works through leg size, theoretical throat, actual throat and effective throat, how the trigonometry links them, what convex and concave profiles do to that link, and how to measure the dimension that actually governs on a real joint. It describes general geometry and fabrication practice. Any number that decides accept or reject on a real job comes from the code or contract documents governing that work.
One weld, several dimensions
A fillet weld sits in the corner of a joint with a cross-section that is roughly a triangle, and a triangle can be described either by its sides or by its height. Leg size measures along the two joint members, which a gauge can reach. Throat measures across the weld from the root to the face, which is where load actually passes and which nothing can touch directly once the weld exists.
Designers work in throat because throat is the section that carries shear. Drawings call out leg because leg is what a fabricator can set up to and an inspector can measure. Geometry connects the two, so a drawing that specifies a leg has implicitly specified a throat, and a weld that meets its leg on a well-shaped profile has the throat the designer intended. Every complication below - convexity, concavity, unequal legs, a corner that is not square - is a case where that assumed relationship has been bent out of shape.
- Leg size: measured along each member from the joint root to the weld toe
- Theoretical throat: the height of the largest right triangle that fits inside the weld
- Actual throat: the shortest distance from the weld root to the weld face as built
- Effective throat: the shortest root-to-face distance with any convexity discounted
Leg size: the number on the drawing
For an equal-leg fillet on a square corner the size is a single number applying to both legs. Geometrically it is the leg of the largest right isosceles triangle that can be inscribed inside the weld cross-section with its two sides lying along the members. That inscribed-triangle definition is not pedantry: it is what makes the size of a hollow weld smaller than its footprint suggests.
On a convex fillet the inscribed triangle reaches the toes, so the leg you can see is the size. On a concave fillet the face curves inward and a straight line between the toes would run through empty air, so the largest triangle that genuinely fits inside the metal is smaller. Two welds with identical toe positions can therefore have different sizes, and the concave one is the smaller of the pair.
This is the reason fillet gauges are sold in convex and concave styles rather than as one universal tool. Each style is cut to find the inscribed triangle for its profile. Reaching for the wrong one does not produce an obviously silly reading; it produces a plausible number that happens to be wrong.
Theoretical throat and where 0.707 comes from
For an equal-leg fillet on a ninety-degree corner, the theoretical throat is the perpendicular distance from the joint root to the hypotenuse of that inscribed triangle. Drop a perpendicular from the right angle of a right isosceles triangle onto its hypotenuse and the result is the leg multiplied by the sine of forty-five degrees. That factor is 0.7071, which everybody rounds to 0.707.
So a quarter-inch equal-leg fillet on a square corner has a theoretical throat of about 0.177 inches, and an eight-millimetre leg gives about 5.7 millimetres. Going the other way, a throat requirement converts to a leg by dividing by 0.707, which is the same as multiplying by 1.414. Both directions turn up in examination arithmetic and both are worth doing without hunting for the factor.
The factor holds only for equal legs on a square corner. Open the included angle out on a skewed tee and the throat shrinks relative to the legs; close it down and the throat grows. Make the legs unequal and the constant disappears entirely. Treat 0.707 as a very common special case rather than as a rule of nature.
Actual throat, effective throat, and convexity
Actual throat is the shortest distance from the weld root to the weld face on the weld as it exists. On a convex fillet that line runs out through the crown, so the actual throat exceeds the theoretical value by roughly the height of the convexity. On a concave fillet the face has been pulled inward, so the actual throat falls short of what the leg positions would suggest.
Effective throat is the portion of that distance the design is permitted to count, which is the shortest root-to-face distance with any convexity discounted. The accounting is deliberately conservative: extra crown metal is weight, heat and distortion rather than strength, because the failure path still runs through the shortest section available to it.
The practical consequence is asymmetric, and it is worth internalising. A convex weld that meets its called-out leg is normally sound on throat. A concave weld whose toes look correct may not be, because the hollow face has removed metal from precisely the section that carries the load. Concavity deserves more suspicion than convexity even though convexity is the more visible of the two.
Unequal legs and skewed joints
Unequal-leg fillets exist because one member can sometimes accept a longer leg than the other, or because the load path favours one direction. The drawing then has to give both legs and has to make the orientation unambiguous, since a fillet with the long leg running vertically is a different weld from the same pair of numbers rotated ninety degrees.
The throat of an unequal-leg fillet is still the perpendicular from the root to the hypotenuse of the inscribed triangle, but that triangle is no longer isosceles and 0.707 no longer applies. It works out as the two legs multiplied together and divided by the length of the hypotenuse. Working that through by hand once stops the relationship feeling arbitrary and makes the equal-leg case obviously a special case of it.
Skewed members behave the same way for the same reason, and codes carry their own provisions for how much skew they will accept and how the throat is credited. The idea to carry out of this section is simply that leg and throat are related by the geometry of the specific joint in front of you, not by a memorised constant.
Measuring the dimension that governs
A gauge only tells the truth when it is seated on sound base metal on both members. The most common bad reading comes from a gauge resting on spatter, on scale, or on the crown of an adjacent bead, which reports a leg the joint does not have. Clean the seating area, set the gauge flat against one member, bring it into contact with the other, and confirm both faces are on metal before reading anything at all.
Match the gauge style to the profile in front of you. A convex-style gauge reads leg on a crowned weld; a concave-style gauge reads the throat-limited size on a hollow one. Using the convex tool on a concave weld gives a wrong answer in the unsafe direction, which is why the pairing gets drilled so hard in training.
Then measure along the weld rather than at one spot. A fillet is rarely uniform over its length, and the dimension that governs the disposition is the one at the worst location, not the one where the gauge happens to sit comfortably. Run the gauge, find the smallest reading, record that value, and note where on the joint you found it.
- Clean the areas where the gauge will seat on both members
- Judge the profile first: convex, flat, or concave
- Select the gauge style that matches that profile
- Seat on both members and confirm metal-to-metal contact before reading
- Run the gauge along the weld to find the smallest reading
- Record the governing value, its location, and the profile you observed
Where this goes wrong in practice
Undersize is the finding everyone expects; oversize is the one that gets waved through. A fillet welded larger than called for adds heat, distortion, consumable cost and very often convexity, and on some connections it shifts the intended failure path into the base metal. Oversize is a legitimate observation to raise even where the governing code does not reject it outright.
The second recurring error is treating throat as something you can measure directly. On a completed fillet you cannot, short of sectioning it. Everything reported about throat on a finished weld is inferred from the legs and the profile, which is exactly why profile conditions carry so much weight and why a hollow weld with acceptable-looking legs still gets written up.
The third is arithmetic under pressure. Converting a leg to a throat, or a required throat back to a leg, is one multiplication in either direction. Being fluent enough that the conversion costs no attention leaves that attention where it belongs, which is on the steel rather than on the factor. That fluency is cheap to buy and it shows up on every joint you measure.