Humans have a limited capacity to analyze odor mixtures with three to four being the maximum [Physiol Behav 46 (1989) 809.]. This study investigates the large loss of information about odor identity that occurs in mixtures and aims to determine the information on which identification and failure to identify is based. In Experiment 1, 14 subjects used a selective attention procedure to identify odorants in stimuli consisting of one to four components. As expected, substantial difficulties were encountered in identifying more than two odorants, and chance level scores were obtained for the group for each of the odorants in the quaternary mixture. In Experiment 2, 21 subjects used a profiling procedure consisting of 146 descriptors to describe the odor qualities perceived in the same stimuli used in Experiment 1. The results indicated that for some odorants, loss of a major characteristic quality occurred even in binary mixtures, but that many of the features of some odorants remained in the quaternary mixture. Comparison of the data from the two experiments indicated that identification of most of the prominent qualities of an odorant was not necessarily sufficient for identification of the odorant in a mixture. In contrast, the loss of some prominent features did not always result in non-identification. A configurational hypothesis of olfaction, analogous to that for facial and object recognition, is proposed to account for the data and the processes underlying odor identification in mixtures.