Judgments of the contingencies between the opinions expressed by three persons in a video-taped group discussion were investigated. Although a purely statistical interpretation of the contingency judgment task was called for by the experimental instruction, the intrusion of non-statistical information in the judgment process was demonstrated: Temporal contiguity (order of speech) and spatial contiguity (eye-contacts, body movements) systematically affected the estimated frequency of agreement among discussion participants. Similar biases were obtained in a memory test for the observed opinion statements which also suggests that intensional information (structural similarity of the discussants' arguments) influenced the cognitive representation of the contingencies. An attentional focus manipulation was also effective; attending to a certain pair of discussants resulted in higher agreement ratings for that pair. The implications of these findings for experiments which use purely statistical models of contingency as a normative criterion are discussed.