What is signal versus noise?

Signal is meaningful, real information; noise is random, irrelevant fluctuation. The model holds that most data is mostly noise, and good judgement means separating the few signals that matter from the static around them.

What is an example of signal vs noise?

Daily stock prices look chaotic (noise) — up and down at random — but over years a clear trend (signal) emerges. Reacting to the daily swings is reacting to noise; the long-run direction is the signal.

How do you separate signal from noise?

Look for patterns that persist across many data points rather than reacting to single ones, zoom out to longer timeframes, and require enough data before trusting a trend. Beware both dismissing real signal and seeing signal in randomness.

Signal vs noise

Information theory & statistics

The signal-to-noise model distinguishes meaningful information (signal) from random, irrelevant fluctuation (noise). Most data is mostly noise, and the core skill of good judgement is separating the few signals that matter from the overwhelming static around them.

How it works

For any stream of data, news, or feedback, ask what is real, repeatable signal versus what is random variation. Resist reacting to noise — single data points, short-term swings, vivid anecdotes — and wait for the pattern that persists across the static.

How to use it

Avoiding overreaction to short-term swings in metrics, markets, or feedback.
Recognising that more data often means more noise, not more clarity.
Distinguishing a real trend from random fluctuation before acting.

Worked example

A daily-checked stock portfolio looks alarmingly volatile — up 2%, down 3%, up 1% — almost all noise. Zoom out to years and a clear upward signal emerges. The investor who reacts to the daily static trades badly; the one who waits for the signal does not.

Where it fails

You can over-filter and dismiss a real signal as noise (a costly false negative), or see signal in pure noise (the clustering illusion). Knowing how much data you need before a pattern is trustworthy is itself the hard part.

The deeper point

It carries a counterintuitive warning: more information often makes judgement worse, not better, because added data is mostly added noise. In an age of infinite data, the scarce skill is no longer gathering information — it’s ignoring almost all of it.

Frequently asked

What is signal versus noise?: Signal is meaningful, real information; noise is random, irrelevant fluctuation. The model holds that most data is mostly noise, and good judgement means separating the few signals that matter from the static around them.
What is an example of signal vs noise?: Daily stock prices look chaotic (noise) — up and down at random — but over years a clear trend (signal) emerges. Reacting to the daily swings is reacting to noise; the long-run direction is the signal.
How do you separate signal from noise?: Look for patterns that persist across many data points rather than reacting to single ones, zoom out to longer timeframes, and require enough data before trusting a trend. Beware both dismissing real signal and seeing signal in randomness.

Regression to the mean Bayesian thinking The map is not the territory Clustering illusion Recency bias

Editorial synthesis © ReadGlobe 2026, drawing on the mental-models tradition (Charlie Munger, Farnam Street) and the primary sources for each model. · Last reviewed 2026-05-29.