Individuals judged how often examples of taxonomic categories had occurred in a study list. An availability hypothesis was tested--that frequency estimates are based on the retrieval of instances. Cued (by category names) recall of the examples served as an index of availability. The hypothesis was confirmed--there were strong positive correlations between frequency judgments and recall (with the influence of actual frequency removed)--given one or more of the following conditions: List instances were not categorized aloud as they were presented; frequency estimation was preceded by cued recall; frequency estimation was delayed by a week. Limitations on availability occurred under other conditions--notably, when individuals, during list presentation, named the categories to which items belonged and received feedback about their categorizations. Under these circumstances, correlations of frequency estimation and recall were often not significantly different from zero, and frequency judgments and recall sometimes reacted differently to changes in independent variables (e.g., frequency judgments of young and elderly subjects did not differ reliably, even though cued recall of young persons markedly exceeded that of elderly subjects).