Abstract: When categorization is uncertain, the inner mechanism of people’s feature induction induced some arguments. In terms of predicting features, Rational Model indicated that people would take all categories into consideration while single-category theory advanced that people only paid attention to the target category. The essential difference between Rational Model and single-category theory is whether people would take the alternative categories into consideration when predicting features. But neither Rational Model nor single-category theory could explain the results of the research by Verde et al in 2005. So Verde et al concluded that the feature prediction in their research might be based on feature association. Therefore, three experiments were designed in this research to investigate how people predicted features when categorization was uncertain.
Experiment 1 replicated the experiment of Verde et al (2005) in Chinese circumstance so that the effect of language would be excluded. In Experiment 2, the frequency of the feature association in order to find out whether categories were considered. Experiment 3 the variable of category was controlled in order to find out whether people’s induction was based on feature association.
Seventy subjects were evenly grouped into experiment 1 and experiment 2, and thirty-four subjects in experiment 3. All of the subjects were individually tested on computers, which controlled list generation, stimulus presentation, and response recording. The procedure comprised five phases: (1) study, (2) a training test, (3) study, (4) a training test, and (5) a final test. During the study phase, all exemplars were shown once. Each exemplar was shown for 4000msec. During the training test phase, subjects should categorize all exemplars into their correct categories. During the final test phase, subjects should decide whether or not the to-be-predicted feature was the one that most likely to appear with the given feature and then to make a yes-no judgment.
Experiment 1 gained the same results as the experiment conducted by Verde et al (2005). By controlling the variable of feature association, Experiment 2 found that there was no significant difference in feature prediction between two conditions, which indicated that people’s induction was not category-based. Via controlling the variable of category, Experiment 3 found that there was significant difference in feature prediction between two feature association conditions, which indicated that people’s induction was based on feature association.
The three experiments illustrate that people are inclined to predict features on the basis of feature association when categorization is uncertain. In contrast to the category-based prediction, prediction based on feature association is an economical strategy, which can be adopted to explain many results of the previous studies

Key words: ingle-category theory, Rational Model, based on categories, based on feature association, feature prediction