We experimentally studied the effectiveness for risk judgment and decision making of two kinds of information which may underlie probability of accident: (a) the relative frequency of similar accidents in the past, and (b) indications of possible ways in which an activity might lead to an accident. Subjects read descriptions of individual choice problems in which paragraphs with different risk information were inserted. Some subjects received Relative Frequency Information (RFI), others Cognitive Scenario Information (CSI), and still others a combination of RFI and CSI. Within each group the quality of the RFI or CSI was varied. Both RFI and CSI affected subjects' accident probability estimates and, although less, their tendency to opt for the riskier alternative. When combined, RFI and CSI still influenced 'riskiness', but the role of RFI was significantly suppressed. The perceived controllability of an activity also appeared to affect risk judgments and decisions. The results are discussed in terms of a fourfold taxonomy of risk.