Password Entropy Explained

What does password entropy mean?

Entropy is commonly expressed in bits and represents the logarithm of the number of equally likely outcomes. If a generator selects each character independently from a pool of N characters for L positions, the theoretical value is L × log2(N).

The word “equally” is essential. The formula describes the generator, not the visual appearance of one result.

Character-pool example

A pool of 62 upper-case letters, lower-case letters and digits used for 16 independent positions has 62^16 combinations. Adding length multiplies the space by 62 for each position. If a service reduces the pool, more length can restore margin.

Use the entropy calculator to inspect the pool, combinations and bits without entering a real password.

Passphrase entropy example

For the EFF long word list, each independent word is one of 7,776 choices. A six-word generated phrase has 7776^6 possible ordered outcomes under the stated process, or about 6 × log2(7776) bits.

Spaces and fixed separators do not add choices. A randomly selected separator adds only its real number of options.

Why human-created passwords do not have simple entropy

People do not choose characters uniformly. They use language, names, dates, keyboard patterns and shared cultural references. Attackers model those choices with dictionaries and probabilistic rules. Counting every character as if it were random therefore inflates the estimate.

A pattern-aware strength checker can identify some known structures, but its output remains an estimate based on a model.

Conditional rules change the search space

If a generator requires at least one character from every selected group, not every string in the full pool is valid. A correct estimate accounts for the excluded strings. This site uses inclusion-exclusion calculations for the displayed constrained space rather than pretending the unconstrained formula still applies.

The generation algorithm rejects candidates that miss required groups, keeping positions random within the accepted policy.

Entropy does not equal crack time

Converting combinations to time requires a guess rate and an attack model. Online authentication should be rate-limited and monitored. Offline guessing after hash theft depends on the password hashing algorithm, cost settings, salt and hardware. Attackers also prioritize likely human choices rather than searching randomly.

Credential stuffing and phishing bypass the search space. State every assumption when presenting a crack-time estimate.

When entropy is useful

• Comparing two uniform random generator policies.
• Choosing word count from a fixed random list.
• Checking the effect of pool size and length.
• Documenting a token-generation requirement.

It is not useful as a precise label for a password with an unknown creation process.

Practical interpretation

Use entropy as one input, then check uniqueness, storage, account rate limits, MFA, recovery and server-side hashing. Prefer a comfortable margin rather than tuning a credential to a dramatic time estimate. The strongest calculation cannot protect a reused or phished secret.

A worked comparison of length and pool size

Suppose one generator uses 62 letters and digits, while another uses 90 printable characters. At the same length, the larger pool has more combinations. Adding one character to the 62-character policy multiplies its space by 62, which can be more useful than adding a few symbols while keeping length fixed.

The comparison remains valid only if selection is uniform and the service accepts the complete value. If a user chooses where symbols appear or repeatedly regenerates until a familiar pattern appears, the simple model no longer describes the selection process exactly.

Min-entropy and conservative reasoning

Security discussions sometimes use min-entropy to focus on the most likely outcome rather than average uncertainty. Human password distributions contain very likely choices, so their conservative strength can be much lower than a pool calculation suggests. A generator designed for uniform output makes the distribution easier to reason about.

Consumers rarely need to calculate advanced entropy measures. The practical lesson is to generate rather than improvise, add a comfortable length margin and avoid claims of exact human-password entropy.

Reporting entropy responsibly

State the alphabet or word list, output length, independent-selection assumption and any group constraints. Separate bits of search space from guesses per second and time. Label average search time correctly and explain that targeted dictionaries, leaked credentials and phishing are different attacks.

A responsible report also says what the number cannot prove. It does not validate service-side hashing, endpoint security, account recovery or whether the value has been reused.