Put a part-used spool on a kitchen scale. It reads 640 g. How much filament is on it?
There are three defensible answers, and they are a long way apart.
- On a cardboard reel: 480 g, about 160.94 m of 1.75 mm PLA.
- On a moulded plastic reel: 425 g, about 142.5 m.
- As a bare refill coil with no reel at all: 640 g, about 214.58 m.
At 45 g a print that is ten prints, nine prints or fourteen prints. The scale was accurate to the gram. The answer was never in the scale.
The number you need is the reel, and you do not have it
A spool sold as one kilogram contains a kilogram of filament. The reel is extra, and its mass is the one input that turns a weighing into an answer. Cardboard reels in the ranges collected here run 140–180 g. Injection-moulded plastic reels run 180–250 g. Those two classes barely overlap, and almost nothing on a spool tells you which one you are holding once the label has scuffed off.
So the honest version of "weigh it and subtract the spool" is: weigh it and subtract a number you are guessing, from a range that is wider than a print. The spool weight reference has the classes and what they look like on a shelf, and what an empty spool weighs is the short answer.
Which way the error points, which is the part that matters
Take the same 640 g reading and resolve it against the extremes of those two ranges. The lightest reel on the list gives 500 g remaining. The heaviest gives 390 g. Same spool on the same scale, and the answer moves by more than two prints.
Now notice the asymmetry, because this is the whole reason to care. Guessing the reel too light makes you believe you have more filament than you do. That error does not show up as a mild disappointment; it shows up as a print running out at ninety percent, on the machine, overnight, with the part unfinished and the material already spent. Guessing the reel too heavy costs you nothing except a slightly early spool change.
The errors are not symmetrical, so your guess should not be either. When you genuinely do not know the reel, assume the heavy end. That is why estimates on this site default to the moulded plastic class rather than the average of everything: the cheap failure is better than the expensive one.
What your printer's own percentage is doing
Machines with filament tracking do not weigh anything. They start from a nominal payload and subtract what the slicer says was extruded, which is a counter, not a measurement.
Counters are excellent right up until reality diverges from them. A spool that was already part-used when it was fitted, a length cut off to clear a jam, a purge tower, a failed print that ran for three hours before you caught it, a spool swapped between two printers — none of those necessarily reach the counter, and every one of them is subtracted from the real filament and not from the estimate. The error only accumulates in one direction, and it is the optimistic one.
A scale reconciles it in ten seconds. It is worth doing before any print you would be annoyed to lose.
The fix takes thirty seconds and lasts the life of the reel
When a spool runs out, do not throw the reel away immediately. Weigh it empty, and write the figure on it in marker.
That single number converts every future weighing on that reel from an estimate into an arithmetic problem. It also, over a few spools, tells you what your usual brands actually weigh, which is more useful than any published table — including the ranges quoted above, because those are class fallbacks for people who have not done this and yours would be a measurement.
If you would rather not wait for a spool to finish, weighing a partial spool has the workarounds, and how much filament is left is the quick version of everything above.
When the answer needs to be metres
Grams are what the scale gives you and what the slicer reports, so grams are usually the right unit. Metres matter in two cases: you are threading a length by hand, or you are trying to decide whether a specific model will fit in what is left.
Converting one to the other needs a density, and density is where estimates go quietly wrong. The figures at the top of this page assume PLA at 1.24 g/cm³. Run the same mass as matte PLA, whose mineral filler makes it measurably denser, and the length drops without anything visible changing. The length and weight converter takes the density explicitly for that reason, and how many metres are on a kilogram gives the baseline for PLA.
The decision this number is really for
Almost nobody wants to know how much filament is left. They want to know whether to start the print.
Which makes the useful output not grams but a margin. Take your best estimate, subtract what the slicer says the job needs, and ask whether the remainder is bigger than your uncertainty. On a guessed reel your uncertainty is the width of the reel range — call it seventy grams of not-knowing on a plastic spool. If the margin is smaller than that, either weigh the reel properly or start a fresh spool, because a run-out on a long print costs the whole print, not the missing few metres.
It is worth being blunt about what that gets wrong when it gets wrong. A print abandoned at the top layer has consumed all of its filament, all of its machine hours and all of its slot, and returns nothing — which is the same arithmetic as any other late failure, and it lands on the biggest, longest, most annoying-to-repeat jobs precisely because those are the ones that empty a spool. Set against that, a partly-used reel put aside for small parts is a trivially cheap insurance policy. If you want a feel for how many jobs a spool should be lasting you in the first place, prints per kilogram is the sanity check.
And if the spool is old enough that you are asking these questions, check it is not tangled before you commit an overnight job to it. Running out is recoverable. A cross-wound snag at three in the morning is not.