The purpose of the present study was to test the predictions derived from Semantic Feature Hypothesis concerning the acquisition of spatial adjective pairs and to discuss the formational processes of semantic space. The subjects were 30 children attending a nursery school in Sendai. According to their ages, they were divided into three groups of 10 each. The mean age for each group was 4; 0, 5; 1, 6; 1, respectively. Two experiments were conducted. In experiment 1, the Ss were presented with 10 spatial adjectives and asked to give their opposites. The 10 adjectives used were big/small, long/short, tall/short, thick/ thin, and wide/narrow. In experiment 2, the Ss were shown a set of four wooden blocks and asked to choose the-one, where the blank was filled with one of the following 6 words: longest/shortest, tallest/shortest, thickest/thinnest. The main results were as follows. (1) The children were aware that a word belonged to a particular semantic space before they had learned the full meaning of the word. (2) “Big/small dimension” was acquired prior to other dimensions. When the children did not know well more restricted adjectives, they tended to substitute “big/small” for them. (3) Unmarked words were acquired no later than marked words. (4) The children of 4 years old tended to regard marked words as complementary sets of unmarked words and the children of 5 years old tended to regard them as being below the average in the relevant dimensions.