The recovery of word-emission ability by Japanese aphasics was examined quantitatively. The scores on a word-emission task were obtained in a series of language tests for aphasia. 17 aphasics were asked to utter names of animals, giving as many as possible within a minute. The task was repeated four times at intervals of 104.1 days on the average, ranging from 27 to 271 days. The score on the word-emission task was estimated by fitting a growth curve to scores on the former three tests. Modified exponential, logistic, and Gompertz curves were used for estimation. All estimated scores by three growth curves were larger than the observed scores. The modified exponential curve which does not have an inflection point produced the most remarkable error. Between two curves which have an inflection point the logistic curve produced less error than the Gompertz curve. The difference in error of estimation was due to the difference in the saturation level of the two curves.