A carbon microphone is shown to have 5 dB less low‐frequency sinusoidal output than a magnetic microphone, both judged to have approximately the same voice output level and tonal quality. Evidence is offered to show that the sinusoidal characterization of a carbon microphone does not properly describe its response to a complex, wide‐band signal having a high peak factor, such as speech. This results from a greater‐than‐unity slope for the input‐output curve of the carbon instrument. An improved characterization for estimating the subjective performance with speech input was obtained with an excitation signal consisting of a sinusoid plus white noise, the latter being filtered from the output. With this technique, the response curves for the two microphones became comparable and reasonably similar. This information is particularly important for testing carbon microphones designed to have a steep input‐output characteristic to exclude farfield sounds.