Self-noise, also called equivalent noise level (EIN), is the electrical noise a microphone produces internally even in complete silence. It is measured in dB(A) — decibels weighted to match human hearing sensitivity. A lower number means a quieter mic.
For most podcasting and voice-over applications, self-noise below 20 dB(A) is excellent and below 30 dB(A) is acceptable. The Audio-Technica AT2020 has a self-noise of 20 dB(A), which is very good for the price. The Rode NT1 achieves an exceptional 4.5 dB(A) — one of the lowest of any microphone at any price.
Self-noise becomes critical when recording very quiet sources — soft acoustic instruments, whispered voice, ASMR — or when applying heavy gain in post-production to increase loudness. With a noisy mic, turning up the gain amplifies the mic's internal hiss along with the signal.
Dynamic microphones generally have higher self-noise than condensers, but this is offset by their lower sensitivity — they need to be closer to the source and are typically used for louder subjects.
In practice, for spoken-word recording in a normal room, self-noise above 28 dB(A) starts to become audible and distracting in quiet moments between sentences.