Subjects were given, as stimuli, either letters or bigrams and asked to think of a word beginning with those stimuli. Words emitted were to be 3, 4, or 5 letters in length. Each of the six groups contained 100 Ss. Subjects were given 60 sec per stimulus to produce a word. Response measures were the number of words emitted, latency of response, and language frequency of emitted words. It was found that responses were fewer, latencies were longer, and word frequencies were lower for bigram than for letter stimuli, and for longer vs. shorter word lengths. It was pointed out that the effective pool of acceptable words is smaller for bigram than for letter stimuli and is probably smaller for longer than for shorter words. The findings with all response measures were interpreted leager terms of a greater difficulty in thinking of words as pool size is reduced.