In the present paper a formula is developed for calculating the hearing loss for speech from an audiogram showing the hearing loss for each of a series of pure tones. The formula is based upon studies of loudness, including the determination of the relative contributions of different frequency regions to the audibility of speech at or near the threshold level. The formula is tested for each of 165 ears involving a wide variety of hearing losses. In every instance an audiogram is available and also an independent observation of the hearing loss for speech. The formula yields a calculated value which generally is in closer agreement with the observation than is the value calculated by the familiar rule of averaging the losses at 500, 1000, and 2000 c.p.s. The agreement is particularly better when the audiogram is not “flat.” A simplified computational rule, indicated by the more complete formula, is found within indicated limits to be almost as reliable as the formula. This simplified rule is to examine the hearing losses measured by means of the audiometer at the three frequencies 500, 1000, and 2000 c.p.s., and to take the average of the two smallest values of loss.