A simple definition, and where law and fact fit in
Science is how we test ideas by evidence: it is a disciplined way of explaining nature that insists every explanation be checked against observation, and revised or dropped if observation disagrees with it. That one sentence covers most of what distinguishes science from opinion, tradition, or intuition — the insistence on checking.
To talk about that process precisely, it helps to have four separate words rather than using "theory" or "fact" loosely for all of them.
A scientific fact is an observation confirmed so many times, by so many independent people, that treating it as reliably true is no longer in doubt — that moths in industrial areas are darker than moths in the countryside, say, once enough surveys agree. A scientific law is different again: it describes a consistent regularity, often as a mathematical relationship, without explaining why the regularity exists. The law of gravity describes how strongly two masses attract each other and lets you calculate it precisely; it does not say why mass bends space that way. A scientific theory is the explanation — the account of the mechanism behind the regularity the law describes and the facts the observations established. Evolutionary theory doesn't just note that species change (that's observation accumulating into fact); it explains the mechanism — variation, selection, inheritance — that produces the change.
Put the three together and the relationship is simple to state: facts are the confirmed observations that need accounting for, laws describe the regular pattern those facts follow, and theories explain why that pattern holds. None of the three outranks the others — a theory doesn't "graduate" into a law, because they answer different questions (why versus what pattern).
An observation of nature needs explaining
Science starts with something odd: a pattern in nature that doesn't yet have an explanation. A biologist notices that a population of moths is darker in industrial areas than in the countryside. An astronomer notices that distant galaxies are redder than expected. These are observations — recorded, repeatable data — and they demand an account of why they happen.
From the observation, someone proposes a hypothesis: a specific, testable explanation narrower than a full theory. The moth researcher might hypothesize that darker moths survive better against soot-stained bark because birds spot the lighter ones more easily. That hypothesis only earns its keep once it generates a prediction someone can check.
Checking it means running experiments and gathering further observations — releasing moths of both colors in sooty and clean woodlands, say, and counting which ones birds take. A single successful test proves little on its own. What moves an idea forward is repetition: the same prediction holding up across different researchers, different populations, different methods.
If an explanation survives that repeated testing — not just once, but across many independent cases — it gains explanatory and predictive power. It starts to account for observations nobody built it to explain, and it starts correctly predicting new ones in advance. At that point the scientific community, not any single researcher, accepts it as a theory: a broadly supported explanation for a range of related phenomena, not a single claim about one case. No hypothesis promotes itself; it is accepted when enough independent groups, publishing and checking each other's work, converge on the same conclusion. The review process that moves a finding from one lab's claim to accepted science is the same process that can later retract a finding when the community decides the evidence doesn't hold — a point worth keeping in mind for the next section.
People hear 'theory' in its everyday sense of a hunch
In daily conversation, "theory" often means a guess — "I have a theory about why the bus is late." That everyday sense is where most confusion about science starts, because people carry it into sentences like "evolution is just a theory" and conclude that scientific theories are unproven speculation.
The confusion collapses once you separate scope from support. A hypothesis is one specific, testable claim about one phenomenon. A theory is a tested explanation that accounts for many separate facts and has survived many independent attempts to disprove it. The two words describe different stages of confidence, not different degrees of guesswork — a theory is, by definition, the opposite of a hunch.
Plate tectonics and evolution by natural selection are usually offered as the clearest examples, because both started as specific hypotheses and were tested against mountains of independent evidence — fossil distribution, seafloor magnetism, genetic sequencing, direct observation of populations changing over generations — before being accepted as theories. Each case shows the same pattern: a theory is not what scientists say when they're unsure; it's what they say once an explanation has survived everything they could throw at it.
This is also where consensus bodies matter. When the National Academies of Sciences, Engineering, and Medicine reviewed the evidence on genetically modified foods and concluded they do not pose special health risks, the conclusion carried weight because it summarized many independent studies rather than one team's result. That's the same mechanism that elevates a hypothesis to a theory: not one lab's confidence, but convergent, checkable evidence reviewed across a field.
| Hypothesis | Theory | |
|---|---|---|
| Scope | One specific, testable claim | Explains many related facts |
| Evidence base | Limited, early-stage tests | Extensive, repeated, independent tests |
| Who accepts it | Proposed by researcher(s) | Accepted by the broader scientific community |
| Can it be overturned | Yes, easily, with one contrary result | Yes, but requires substantial contrary evidence |
New observations conflict with an accepted theory
Sometimes new data don't fit an accepted theory, and that's not a crisis — it's how science is supposed to work. The mismatch is first confirmed by replication: other researchers rerun the observation or experiment to check it wasn't a measurement error or a one-off anomaly.
If the conflict holds up under independent review, the theory gets modified, narrowed in scope, or in rarer cases replaced by a better-supported explanation. Newtonian mechanics wasn't thrown out when relativity arrived — it was narrowed to the domain where it still works extremely well (ordinary speeds and masses), while relativity extended the explanation to extreme speeds and gravity that Newton's equations couldn't handle.
Whatever replaces an older theory has to keep explaining everything the old one explained correctly, and it still has to account for the same accumulated facts and obey the same underlying laws those facts fit. A successor theory doesn't get to ignore the facts its predecessor already accounted for; it has to absorb them and add the new cases on top. That requirement is why revision counts as progress rather than failure — the older theory wasn't simply "wrong," it was incomplete, and science's job is to keep narrowing that incompleteness.
The same institutional check that promotes a hypothesis can also pull a published finding back out of circulation. A science journal retracted a study about a disputed form of life after other researchers reviewed the methods and evidence and found them unconvincing — against the original authors' wishes. That episode is a sharper version of the same process described above: publication is not the end of scrutiny, and a claim stays open to correction for as long as other scientists can check it.
A claim is examined for whether it predicts anything observable
The single question that sorts science from opinion is simple: does the claim predict something that could, in principle, turn out false? If no observation could ever contradict a statement, it isn't wrong exactly — it's just not science. It stays an opinion, a belief that doesn't have to answer to evidence either way.
That property has a name in the field: testability, sometimes called falsifiability. A testable claim stakes something on the outcome of an observation or experiment. "This medication reduces fever within six hours" is testable — a trial could show it doesn't. "This arrangement of furniture brings good fortune" isn't, because no measurement could ever settle it either way.
If a claim is testable, it can proceed through the ordinary path: proposed as a hypothesis, checked by experiment and observation, and — if it survives enough independent tests — eventually accepted as part of a broader theory. If it isn't testable, none of that machinery applies, no matter how confidently it's stated or how many people believe it.
That one question is worth keeping as a working habit: before treating a claim as established science, ask what observation would have to occur for the claim to be false. If there's a clear answer, it belongs to the realm science can examine. If there isn't, it's an opinion — which may be reasonable, useful, or widely shared, but is not something evidence can confirm or overturn.