Seed Germination Calculator

In short Two questions come out of a packet of seed: what proportion of it is alive, and how many should go in each cell. The first is a count — ten seeds on damp kitchen paper in a warm place, counted after the number of days on the packet — and the second is probability, which is where the intuition usually goes wrong. Seeds fail independently, so the failures MULTIPLY. At 80% germination one seed leaves a cell empty a fifth of the time. Two seeds both fail only 4% of the time, so the cell fills 96% of the time; three seeds get to 99.2%. That is why two per cell is not twice as good as one — it is five times better measured in empty cells, and each extra seed cuts the empty ones by the same factor again. It also explains the diminishing returns: past three seeds at a decent germination rate you are no longer buying full cells, you are buying thinnings. Sowing for a target number of plants works the same way. Losses multiply rather than adding: 80% germination followed by 90% surviving to planting out is 72%, not the 70% you get by taking both losses off one. For 24 plants that is 34 seeds, where the 70% figure would have you sow 35. This page does both, and shows the expected thinnings alongside, because those are a real cost in bench space and in expensive F1 seed even though they are free in cheap lettuce. Two things worth being honest about. A ten-seed test is a small sample, and the page now says how small: eight of ten is consistent with a true rate anywhere from 49% to 94%, at 95% confidence. That interval is computed from your own test rather than quoted, so it tightens as you test more seeds — fifty seeds at the same 80% narrows it to 67% to 89%. Use a ten-seed test to decide whether to sow thickly, not to argue with the supplier. And germination falls with age at very different speeds depending on the species: onion, parsnip and leek are usually finished after a year or two, while tomato, cucumber, brassicas and squash keep for five years or more in a cool dry place. Storage conditions matter far more than the date printed on the packet, so when in doubt run the towel test rather than reading the year.