Point a telescope at a star and record its brightness, over and over, for months. Most of the time nothing happens.
If a planet's orbit happens to lie edge-on from our point of view, it slides across the star's disc once per orbit and blocks a slice of the light. The star dims by a fraction of a percent, stays dim for a few hours, then recovers.
One dip proves nothing — a cloud, a sunspot, an instrument glitch all look similar. What makes a detection is the dip returning on a strict schedule, with the same depth and the same duration every time.
The arithmetic
Depth of the dip
δ = (Rp / R*)²
The fraction of light blocked is just the ratio of the two discs' areas. This is why the method measures a radius and not a mass — it never touches the planet's weight.
Chance the orbit is edge-on enough
Ptransit ≈ R* / a
a is the orbital distance. Close-in planets transit far more often, which is the single biggest bias in the whole catalogue.
Jupiter and Earth, seen from outside
Jupiter across the Sun
(69,911 / 696,340)² = 0.0101
a 1.0% dip — easy
Earth across the Sun
(6,371 / 696,340)² = 0.000084
0.0084% — 84 parts per million
Transit odds, Earth's orbit
696,340 / 149,600,000
0.47% of viewing angles
That 84 ppm is why finding an Earth twin needs a space telescope: the atmosphere alone makes the ground-based measurement wobble by more than the signal.
What it does well
Finds small planets — the only method that routinely reaches Earth-size
Scales to millions of stars at once, which is why it dominates the catalogue
The same dip reveals the atmosphere when the star's light filters through it
Where it cannot help
Needs the orbit almost exactly edge-on — misses the overwhelming majority of planets
Gives radius but never mass, so it cannot tell rock from gas on its own
Starspots and stellar flickering mimic shallow dips; false positives are common
Landmark discovery
HD 209458 b (1999) — the first planet ever seen to transit, and the first with a detected atmosphere.