The cold you feel at the glass is usually not a leak, and the leak is usually somewhere you cannot feel. Four tests that sort them before you buy caulk.
Cold air at the window, in a house that is otherwise warm. Before you buy anything:
And if the glass itself is what is fogging or wet on the inside, that is a different diagnosis on a different page — moisture, not air.
The call comes in January, usually after the first week the daytime high does not get above zero.
"There's a draft at the living-room window. I can feel it from the couch. I want it sealed." The homeowner has already been to the hardware store. On the kitchen counter there is a tube of silicone, a roll of foam tape, and a box of the plastic shrink-film kit that goes on with a hair dryer. Three fixes for three different problems, bought for one symptom.
I put the back of my hand near the glass. Cold. I put it near the bottom corner of the sash, where the frame meets the sill. Nothing. I hold a lit stick of incense six inches from the meeting rail and the smoke goes straight up. This window is not leaking air. It is leaking heat, and the two are so different that the fix for one does nothing for the other.
This is the page I wish I could hand people at the hardware store. Four causes, four tests, and only one of them is fixed by what is in the bag.
Start with the thing that fools everyone, including cleaners who should know better.
A pane of glass on a 20°F day has an inside surface temperature well below the room's. Room air touching it cools, gets denser, and falls. What arrives at your ankle is a slow, steady column of cold air sliding down the glass and spilling across the floor. It feels exactly like a breeze from outside. It is not. No air has crossed the wall. Sealing every joint on that window with a torch and a welder would not change it, because the mechanism is heat leaving through the glass, not air coming in around it.
You can prove this to yourself in thirty seconds. Hold the incense stick a hand's width from the center of the glass. If the smoke drifts down the pane and toward the floor, you are watching convection off a cold surface. If it drifts sideways, or is pushed away from the window, or gets sucked toward a joint, you have found air movement, and now it is worth looking for where.
The distinction matters because the fixes live in different aisles. Convective loss off the glass responds to one thing: putting a still layer of air between you and the cold surface. That is what the shrink-film kit does, what an interior storm panel does, what a heavy curtain drawn to the sill does. It is not what caulk does.
Windows account for roughly a quarter to a third of a house's heating and cooling energy use, and that figure includes both mechanisms.1 In a modern double-pane unit with intact weatherstrip, the leakage share is small and the conduction share is most of it. In a 1920s single-pane double-hung with a century of paint on the parting bead, the proportions flip. Which one you have decides which aisle.
Real air leakage — outside air crossing the wall into the room — has four addresses on a window. None of them is the glass.
The sash-to-frame joint. This is the joint that moves every time the window opens. It is sealed by weatherstripping: a pile strip in the channel of a slider, a compression bulb on a casement, a spring-metal or vinyl fin on a double-hung. Weatherstrip wears. Pile mats down, bulbs take a set and stop rebounding, fins crack in the cold. When it is gone, the window closes but does not seal.
The frame-to-wall joint. The window unit sits in a rough opening cut into the wall, and that opening is larger than the unit — by a quarter inch to a half inch on every side, so the installer can plumb and shim it. That gap runs the whole perimeter of the window, and it is hidden on the inside by trim and on the outside by siding, brick mould or a flange. On the day of installation it was supposed to be filled with low-expansion foam or backer rod and sealant. Very often it was stuffed with fiberglass, which stops nothing, or left empty. The DOE's own weatherstripping guide flags this gap as often the major source of air leakage on a window, and it is the one you cannot see.
The meeting rail and the lock. On a double-hung, the two sashes overlap at the meeting rail, and the cam lock is what pulls them tight against each other and against the weatherstrip. A window that is closed but not locked is not sealed. A window whose lock has been painted over, or whose sashes have racked so the lock no longer engages, has a continuous gap across the middle of the window. Same on a casement: the latch is what compresses the bulb.
The glazing bead. Where the glass meets the sash. On an old window this is putty, and putty that has cracked and pulled away from the glass is an open joint the full perimeter of the pane. On a newer window it is a rubber gasket or a vinyl bead, and those fail too, less often and less visibly.
Four joints. Two of them move and two of them do not. That single fact decides what you are allowed to put on them.
The rule that every energy-office guide states, and that the hardware-store bag violates, is this: weatherstripping seals the components that move; caulk seals the components that do not.
The joint that does not move is frame-to-wall. On the exterior, caulk the head and the two jambs where the frame meets the siding or the brick mould. Do not caulk the sill — I have written a whole page on why the sill is a drain and what happens when it is sealed shut. On the interior, the frame-to-wall joint is behind the trim, and the material for it is not caulk at all but low-expansion foam in the gap itself, with the trim taken off to reach it.2
The joints that move are the sash-to-frame joint, the meeting rail and, functionally, the glazing bead. Caulk on any of these is a mistake with a long tail. Caulk the sash to the frame and you have a window that can no longer open, that is still leaking at the meeting rail, and that will tear the paint off the stop when someone eventually forces it. Caulk the glazing bead on an old sash and you have glued the glass into a joint that needs to be reglazed, not sealed.
This is the point at which the tube of silicone on the counter goes back in the bag. Its one legitimate job on a window is the exterior head and jambs, and if those were caulked when the house was sided they are probably still fine.
The order below is the order that rules things out fastest. Pick a cold, windy evening; a still afternoon tells you nothing.
One: the incense test at the glass. Light a stick, hold it a hand's width from the center of the pane. Smoke that slides down the glass is convection. Smoke that moves sideways or is pushed off the window is air leakage. If it is convection and only convection, stop here — you are in the curtain-and-storm-panel aisle and nothing below applies.
Two: the incense test at the joints. Move the stick slowly around the perimeter of the sash, then across the meeting rail, then around the trim where it meets the wall. Watch for the smoke to bend, jitter, or get pulled toward a joint. Mark each spot with a piece of painter's tape. You are mapping, not fixing.
Three: the dollar-bill test. Open the window, lay a bill across the sill, close and lock the window on it. Pull. If it slides out without resistance, the weatherstrip at that point is not compressing — worn, missing, or the sash is not being pulled tight. Repeat at each side and at the meeting rail. This test is the one that separates "weatherstrip is dead" from "lock is not engaging," because you do it once locked and once merely closed. If the bill holds when locked and slides when closed, the weatherstrip is fine and the problem is a lock that is not being used or not reaching.
Four: the trim test. Take the flashlight and, with the room lights off, have someone shine it from outside along the window's edges while you look from inside at the joint between the trim and the wall, and between the trim and the frame. Light coming through means an open rough-opening gap. No light does not clear it — the gap can be blocked by fiberglass that stops light and passes air — but light is a confirmed find.
Between them, the four tests sort the window into one of five bins: convection only; sash weatherstrip; lock or racked sash; rough opening; glazing. Each bin has one fix, and none of the fixes overlap.
Convection only. Reduce the heat loss through the glass. In order of cost: a heavy curtain that reaches the sill and is drawn at dusk; the shrink-film kit, which is a disposable interior storm; a rigid interior storm panel, magnetic or compression-fit, which is a reusable one; an exterior storm window; and, at the far end, a replacement unit with a lower U-factor. The NFRC label on any new window lists U-factor, solar heat gain and, optionally, an air-leakage rating — they are separate numbers because they are separate problems, and the U-factor is the one that addresses this bin.
Sash weatherstrip. Replace it. Match the profile: pile for channels, bulb for compression, V-strip or spring metal for the sides of a double-hung. Clean the channel first; new pile on old grit mats down in a season. Adhesive foam tape is the hardware-store default and it is the shortest-lived option on the shelf — fine for a fixed window or a sash that is never opened, wrong for a sash that moves daily.
Lock or racked sash. If the lock does not reach, adjust or replace the keeper; a cam lock with an offset keeper takes up most of a sash that has settled. If the sashes are racked — you can see the meeting rail is not parallel to the sill — the window needs to be squared, and on an old wood double-hung that is a cord-and-weight job, not a caulk job. Then use the lock. A closed, unlocked double-hung in January is an open window.
Rough opening. This is the one that pays. Score the paint at the trim, pull the casing carefully, and look at the gap. Empty or fiberglass: pull the fiberglass out, fill with low-expansion window-and-door foam, let it cure, trim flush, re-hang the casing. This is the largest single leak on most windows and it is invisible until the trim comes off, which is why the caulk tube never touches it.
Glazing. Old putty that has cracked and lifted: reglaze. It is a real skill, and on a house full of original sash it is worth learning or paying for once, because a properly glazed and painted pane is a thirty-year seal. A failed gasket on a modern unit: usually a manufacturer part, and worth a call before you improvise.
This is a cleaning site, and there is a reason the draft page lives here. Every one of these joints is a place the cleaner's hands go, and two of them are places we routinely make worse.
The first is the sash channel. A pile weatherstrip that has spent a season collecting grit from a dirty track is a pile weatherstrip that has stopped sealing, and a cleaner who runs a wet cloth down the channel and leaves the grit in the pile has just packed it tighter. The tracks page covers the vacuum-first order for exactly this reason. Vacuum the channel, then wipe; never the other way.
The second is the meeting rail. Cleaners lift sashes. When the job is done, the sash goes back down and the lock stays open, because the lock is not the cleaner's job and the customer will do it. The customer does not do it. I have started closing and locking every double-hung I open, and telling the customer I did, because an unlocked meeting rail on a north wall is the most common "draft" I get called back for.
The third is one to notice, not to touch: putty. If a cleaner can see the glazing has pulled from the glass — a dark line, a lifted edge, a crumb of putty in the track — that goes in the notes and in the conversation at the door. It is not a cleaning problem and we do not fix it. But we are the only trade that looks closely at every pane on the house twice a year, and a glazing failure caught before the water gets behind it is a repair; caught after, it is a new sash.
Test at the glass first: if it is convection, you are done diagnosing and you want a curtain or a storm, not a sealant. If it is air, map the joints with smoke and tape. Dollar-bill the weatherstrip, locked and unlocked, to separate dead strip from a lock that is not doing its job. Flashlight the trim for the rough-opening gap, and assume the gap is there even if the light does not show. Then fix the one bin you are actually in — weatherstrip, lock, foam, or glaze — and leave the tube of silicone for the exterior head and jambs, if they even need it. The window that still feels cold afterward, with the smoke going straight up at every joint, is a window that is doing what glass does in January. The fix for that one is a curtain, and it is the cheapest fix on the page.
On the 25–30% figure. The Department of Energy and the ENERGY STAR program both put windows at roughly a quarter to a third of residential heating and cooling energy use. That number is the whole window — conduction through the glass and frame plus air leakage through the joints — which is exactly why 'my windows are drafty' is not one problem. The share that is leakage, and therefore sealable, is the smaller part in a modern double-pane unit and the larger part in an old single-pane sash. ↩
Why the expanding-foam warning is on every DOE-derived page. High-expansion polyurethane foam develops pressure as it cures. Sprayed into the rough-opening gap it pushes the jambs inward; a double-hung that was free the day before now binds at the meeting rail, and a casement stops latching. The window-and-door formulations are low-expansion for that reason. If the gap is already foamed and the sash binds, the fix is to cut the foam back, not to plane the sash. ↩
Regional contributor covering the Upper Midwest for Window Washing Guide. Coverage under this byline addresses Twin Cities residential routes, ice-dam meltwater chemistry, and lake-country cabin work, researched with the Giordano Inc. editorial team from practitioner interviews and published water-quality and trade references.