A discrimination test says whether the change shows
Sometimes you do not need the whole product described. You need a binary answer: whether the change you made is perceived, or goes unnoticed.
It is the shortest question in sensory work, and it is settled before you commit to a formula. There are four designs that answer it, and the industry calls them by name: triangle test, duo-trio, tetrad and 2-AFC.
When all you need to know is whether it shows
Two one-off changes, both with the same shape: something has moved in the product and nobody should be able to tell. Describing the whole product to find that out is expensive, and the answer arrives late.
You change an ingredient or a supplier
A raw material whose price has risen, an additive coming off the list, another origin for the same component. The question is not how the product turns out: it is whether anyone will notice.
If what is on the table is a stepwise reduction of sugar, salt or fat, the discrimination test is one piece inside a longer project.
Reformulation →You change a process or a plant
An identical formula can give a different product on another line, with other equipment or with another process time. The test settles it once, before the change is approved, rather than after consumers settle it for you.
If that comparison against the control has to be repeated lot after lot, it is no longer a one-off test: it is a monitoring programme.
Sensory shelf life →
The number of tasters follows the decision, not the other way round.
A discrimination test is designed by fixing in advance how much difference you are willing to let through and how sure you want to be. A test with too few tasters almost always concludes that there is no difference, and that is not a reassuring result: it is a test that could not have detected one.
The difference you let through
A change can be real and irrelevant at the same time. Deciding how much counts as irrelevant is a business call, not a laboratory one, and it is made before the first sample is served.
How sure you want to be
The more certainty you ask for, the more tasters it takes. That is the real price of the test, which is why it is worked out at the start and not once there are results to interpret.
Four designs, and the choice is not neutral
The design is chosen by the question in front of you; the number of tasters, by the decision above. All four answer the same thing — it shows or it does not — but they do not ask the same of the taster, and they do not perform the same way on the same sample.
A badly chosen design can miss a difference that is really there, and that result reads as a green light.
| Design | What is served | What is asked | When to choose it |
|---|---|---|---|
| Triangle test | Three coded samples: two identical and one different. | Which of the three is the odd one out. | You do not know which attribute the difference might sit in. It is the starting point. |
| Duo-trio | A reference and, separately, two samples. | Which of the two matches the reference. | There is a clear control: the current product against the reformulated one. |
| Tetrad | Four coded samples. | How they group into two pairs. | You need more statistical power with the same number of tasters, at the cost of more sensory load per session. |
| 2-AFC directional | Two samples and a directional question. | Which is sweeter, more intense, more bitter. | You already know which attribute moved and only need to confirm by how much. |
How far the answer goes
The binary question is settled, together with the confidence the test was designed for and the number of tasters who actually answered. With that you sign the change off or send it back to formulation, and the conversation stops being a contest of opinions.
What is not settled is what the difference consists of. A discrimination test says there is one, not which one or how big. As soon as the answer is that it does show, the useful question becomes in which attribute and by how much.
Have a change to validate?
Tell us what you changed and how much difference you would be willing to accept. We tell you which design fits and how many tasters it takes for the result to mean something.