A new pitch extraction algorithm has been developed which is designed to be robust against the degradations encountered in telephone speech. The algorithm determines the pitch from the spacing between harmonics in the first formant region, using an iterative technique whereby larger peaks are weighted more heavily. The desired spectral region is obtained with better than 7 Hz resolution using only a 128-point FFT, after simple digital signal processing tricks. The buzz-hiss decision relies solely on the continuity of the pitch track, because standard indicators are often unreliable in the presence of noise and distortion. The algorithm has been incorporated into a real-time linear prediction vocoder implemented in the Lincoln Digital Voice Terminal, and is being evaluated on speech that has first been processed through a digital telephone simulator. [Work supported by ARPA.]