Understanding Base64 Padding (=)

Base64 processes input in groups of 3 bytes (24 bits), producing 4 Base64 characters per group. But what happens when the input length is not a multiple of 3? This is where padding comes in.

How padding works:

• If the input has a number of bytes divisible by 3: no padding needed. Example: "ABC" (3 bytes) encodes to "QUJD" (4 characters, no =).
• If there is 1 remaining byte: the encoder produces 2 Base64 characters and appends ==. Example: "A" (1 byte) encodes to "QQ==".
• If there are 2 remaining bytes: the encoder produces 3 Base64 characters and appends =. Example: "AB" (2 bytes) encodes to "QUI=".

The = character is not part of the Base64 alphabet. It is purely a signal that tells the decoder how many bytes to discard from the final group.

Why padding exists: Padding ensures that every Base64 string has a length that is a multiple of 4. This was important in early implementations that processed Base64 in fixed-size blocks and needed to concatenate multiple Base64 strings without ambiguity. If you concatenate two padded Base64 strings, a decoder can process them sequentially without confusion.

Can you omit padding? Yes, in many modern contexts. The decoder can calculate the expected padding from the string length: if length % 4 == 2, add ==; if length % 4 == 3, add =; if length % 4 == 0, no padding needed (length % 4 == 1 is always invalid).

JSON Web Tokens (JWTs) always omit padding. Many URL-safe Base64 implementations omit it as well. However, some strict decoders (including atob() in browsers) require padding and will throw errors without it.

Practical advice:

• When encoding: include padding unless the consuming system explicitly expects it omitted.
• When decoding: be prepared to add padding back if it is missing. A simple one-liner handles this:

function addPadding(base64) {
  return base64 + "=".repeat((4 - base64.length % 4) % 4);
}

Common mistake: Treating = as part of the encoded data. The padding is metadata about the encoding process, not a representation of any input byte.