The mean square noise current per unit frequency range as a function of frequency, or ``power spectrum,'' of the shot noise produced by a secondary‐emission multiplier tube is calculated. A general expression for it, (3), is set up and then applied to the RCA 931 family of photo‐multiplier tubes. These tubes are of interest as relatively intense sources of random noise of constant intensity up to high frequencies. When an electron enters the collecting space, it initiates a pulse of current in the anode circuit. In the 931 operated with 100 volts per stage, the pulse has a total duration of about 6×10−10sec. and a half‐width of about 1.5×10−10sec. The electrons descended from the same photon do not enter the collecting space simultaneously but are spread out over a time of the order of 6×10−10sec. Because of the non‐zero duration of the pulses and the spread in their time of occurrence, the noise spectrum is essentially flat only up to several hundred Mc, falling rapidly at higher frequencies. The paucity of data and the approximations made in the calculation make our curve of noise outputversusfrequency significant only as regards order of magnitude and general trend. It does seem safe to predict that the noise intensity ought to be constant from zero up to about 100 Mc, begin to fall off appreciably between 100 Mc and 1000 Mc, and become very weak at higher frequencies. Partial confirmation is provided by unpublished measurements of Jastram at the Radio Research Laboratory, Harvard University, which show that at 30 Mc the noise output is still about the same as in the range 0 to 5 Mc. There is also some experimental indication that the output is about the same at 100 Mc and that it is much smaller at higher frequencies.