No testimony can rationally establish a miracle, because it is always more probable that the witnesses are mistaken or lying than that a law of nature was suspended

The argument at its strongest

The classic argument here is David Hume's argument "Of Miracles" in An Enquiry concerning Human Understanding. In its strongest form, it is not the crude claim that miracles are impossible because "science has disproved them." Nor is it merely the claim that religious people are gullible. Its more serious form is an argument about evidence, probability and rational belief.

No testimony is sufficient to establish a miracle, unless the testimony be of such a kind, that its falsehood would be more miraculous, than the fact, which it endeavours to establish.

The thought is this. A miracle is usually defined as a violation, suspension or exception to a law of nature. A law of nature is not a casual generalisation. It is a regularity supported by an immense and stable body of experience: dead people stay dead, celestial bodies do not split on command, water does not become wine without ordinary causal processes, and so on. The evidence behind such regularities is not one person's report, but the accumulated experience of humanity under countless circumstances.

Testimony, by contrast, is fragile. People misperceive. They exaggerate. They lie. They remember events in ways shaped by expectation and communal pressure. They are attracted to wonder, status, identity and religious significance. Reports pass through retelling, translation, apologetic use and institutional preservation. Even sincere witnesses can be wrong. Therefore, when we compare two hypotheses - "a law of nature was suspended" and "the report arose through error, deception, legend or misunderstanding" - the second will normally be much more probable.

J. L. Mackie gives a recognisably Humean version of this in The Miracle of Theism. Antony Flew and many later sceptics make similar points. Among historians, a related but narrower position appears as methodological naturalism or methodological restraint: academic history normally works with publicly assessable sources, human actions, social conditions, material evidence and causal analogies with ordinary experience. This is not a proof of metaphysical naturalism. It is a rule about what can be established by the discipline's ordinary methods. On that rule, a historian may conclude that some disciples sincerely believed they saw a risen teacher, that a community preserved a miracle tradition, or that sources report an extraordinary event; but the further claim that "God raised him" or that "a supernatural sign occurred" is usually treated as a philosophical or theological inference rather than a standard historical conclusion.

So the argument is not simply emotional hostility to religion. It is an evidential challenge. If the prior probability of a miracle is extremely low, and if testimony is often unreliable, then miracle testimony must be extraordinarily strong before belief is rational. The central claim is that actual religious testimony never reaches that level.

Where the argument has real force

This argument has considerable force against many popular appeals to miracles. Online and devotional claims often move too quickly from "many people reported it" to "therefore it happened." That is not enough. Group testimony can be dependent rather than independent. A crowd can share an expectation, a rumour, a ritual context or a social incentive. Ancient sources may be late, partisan, theologically shaped or difficult to cross-examine. Even modern miracle claims often lack medical records, controlled observation, careful exclusion of alternatives or access to the alleged witnesses.

The argument is also right to insist on comparative probability. The question is not whether a miracle report has some evidence. It often does. The question is whether the evidence is better explained by the miracle than by non-miraculous alternatives. In many cases, mundane explanations are not ad hoc evasions. Misdiagnosis, psychosomatic recovery, embellishment, fraud, symbolic storytelling, collective expectation and later legendary development are common enough that they must be taken seriously.

There is also a legitimate disciplinary point about history. Historians ordinarily cannot run controlled experiments on unique past events. They work with texts, memories, inscriptions, artefacts and probabilities. Unique events can certainly be studied historically, but explanations are usually assessed by comparison with known patterns of human behaviour, social causation and material evidence. If a proposed explanation invokes a supernatural agent, it may be difficult to test, compare or constrain in the usual historical way. Thus a historian may reasonably avoid saying, in a professional historical conclusion, that a supernatural event occurred. This restraint is not necessarily prejudice; it may be a rule of method.

The main problem with the sweeping version

The difficulty comes when the argument becomes universal: no testimony can rationally establish a miracle. That stronger claim is much harder to defend.

First, Hume's appeal to "uniform experience" can risk circularity if it is read too strongly. A careful Humean need not be saying, "miracle reports are false because miracles never happen, and we know miracles never happen because all miracle reports are false." That would beg the question. The stronger version is that our direct and repeatable experience of ordinary regularities is extremely weighty, while miracle reports usually arrive through fallible chains of testimony. Still, if alleged miracle testimony is excluded in advance from what counts as human experience, then the conclusion has been built into the method. The real question is whether any reports are credible enough to count as evidence rather than as automatically disqualified exceptions.

Second, testimony is not uniformly weak. Testimony is one of our main sources of knowledge. Most of what we believe about geography, history, science and current affairs depends on testimony. Of course testimony can fail, but its force varies with conditions: number of witnesses, independence, competence, proximity to the event, absence of motive to deceive, hostile corroboration, documentary control, physical traces and the cost of maintaining the claim. A blanket dismissal of testimony proves too much.

Third, a Bayesian analysis does not support the claim that testimony can never overcome a low prior probability. John Earman, in Hume's Abject Failure, argues that Hume's argument is not a successful general theorem. If a miracle has a very low prior probability, then the evidence required will be very strong. But "very strong" is not the same as "impossible." Multiple independent and highly reliable lines of evidence can, in principle, overcome a low prior. We may still think actual miracle claims fail; that is a case-by-case judgement, not a proof that testimony could never suffice.

Fourth, the definition of miracle as a "violation" or "transgression" of a law of nature is philosophically contested. Hume himself frames a miracle in terms of a law of nature being transgressed by divine volition or by an invisible agent. But not all philosophers of religion, and not all philosophers of science, understand laws as metaphysical barriers that could not admit an exceptional cause. Some theists define a miracle not as a contradiction inside nature, but as an unusual event brought about by divine agency. On that view, laws describe how natural causes operate when left to themselves; they do not by themselves settle what could occur if an additional personal cause acted within the world. One need not accept that account to see that Hume's argument depends partly on a disputed concept of law. If laws are descriptions of stable patterns rather than exceptionless metaphysical necessities, "violation" may be the wrong category.

The strongest Humean reply

The defender of Hume can answer that these objections are mostly theoretical. Yes, testimony can be strong. Yes, Bayesian probability allows low priors to be overcome in principle. Yes, "law of nature" is philosophically complex. But when we look at actual miracle reports, especially ancient religious ones, the evidence is nowhere near strong enough.

This reply has weight. Reports of resurrections, prophetic signs or cosmic portents often come to us through communities for whom the event already has deep religious meaning. That does not by itself show that the reports are invented or insincere, but it does mean they are not neutral laboratory records. We rarely have direct access to the witnesses. We often cannot establish independence. In many cases the sources are later than the alleged event, and they may be shaped by preaching, memory, liturgy, communal identity and apologetic needs. Alternative explanations may not be individually provable, but the combined space of error, embellishment, misperception, symbolic narration, fraud and later legendary development may be much larger than the specific hypothesis that a supernatural interruption occurred.

The Humean may also say that background knowledge matters. We know many cases of deception, hallucination, misremembering and legend formation. We do not have comparably secure, independently verified cases of violated natural law. So even if testimony could in principle establish a miracle, the bar is extremely high in practice. The sceptic does not always need to identify one specific natural alternative with certainty; in probabilistic reasoning, a disjunction of ordinary alternatives can outweigh a single extraordinary hypothesis. But that disjunction still needs discipline: it should be grounded in known mechanisms of error and transmission, not used as a vague escape hatch whenever testimony is inconvenient.

This is a serious reply. It is especially strong against claims that rest on thin, late, dependent or sectarian testimony. It also explains why many responsible historians, including those not personally hostile to religion, refrain from writing supernatural conclusions into historical reconstructions.

What the argument does not establish

Still, the conclusion must be kept precise. The Humean argument does not establish atheism. It does not show that God does not exist. It does not show that miracles are impossible. It does not even show that no testimony could ever rationally support a miracle. At most, in its strongest defensible form, it gives a demanding evidential standard and argues that known miracle reports fail to meet it.

It also does not show that naturalistic history is the whole of rational inquiry. A discipline may adopt methodological naturalism for practical reasons without proving metaphysical naturalism. For example, a court, a laboratory and a history department use different rules of evidence. A historian's reluctance to certify a miracle as a historical conclusion may be methodologically appropriate, but it does not by itself settle what a philosopher, using a broader set of considerations, may rationally believe.

Nor should sceptics overstate science. Science has established many regularities with extraordinary success. But an open research question should not be treated as proof of naturalism, and a current inability to explain an event naturally would not automatically make it supernatural. Both sides must avoid that mistake. "We do not yet know the natural explanation" is not equivalent to "there is no natural explanation." Equally, "science usually finds natural explanations" is not equivalent to "a supernatural explanation is impossible."

A balanced verdict

The popular slogan - "miracles cannot be believed on testimony because laws of nature are always more certain" - is too quick. It relies on a sweeping contrast between strong laws and weak testimony that is not always justified. Testimony varies in quality, and probability theory does not rule out, in principle, testimonial evidence strong enough to support a very surprising event.

But the more careful Humean challenge remains powerful. A miracle claim bears a heavy evidential burden. The more an event conflicts with stable background knowledge, the stronger and cleaner the evidence must be. Many religious miracle reports, especially those transmitted through ancient communal memory and theological interpretation, do not obviously satisfy that burden by ordinary public standards.

So the argument succeeds as a caution against easy appeals to miracle testimony. It supports scepticism toward many alleged miracles and explains why naturalistic historical method resists supernatural conclusions. It fails, however, as a universal proof that testimony could never rationally establish a miracle, and it fails as a direct argument for atheism. The most it establishes is conditional and modest: if a theistic case depends entirely on contested miracle testimony, then it must meet an exceptionally high evidential standard, and many such cases may not meet it.