Two semantic theories of word learning, a perceptual complexity hypothesis (H. Clark, 1970) and a quantitative complexity hypothesis (E. Clark, 1972) were tested by teaching 24 preschoolers and 16 college students CVC labels for five polar spatial adjective concepts having single word representations in English, and for three having no direct translation into English. Neither theory was supported: polarity, verticality, and number of features were inconsistently related to speed of learning. The failure to replicate earlier studies is attributed to the fact that conceptual dimensions were unconfounded in the present experiment. At the end of the study, the preschoolers' understanding of the real English words was assessed in a multi-trial situation. The likelihood that any given concept would be responded to correctly was reasonably predicted by the frequency of occurrency of that word in the language environment. (Author)