The role of verbal encoding in odor recognition memory was investigated using odors of low familiarity to subjects before the experiment began. The experimental procedure included two phases--odor learning (first phase) and odor memory testing (second phase)--separated by a delay of 7 days. Five experimental conditions were established: three conditions of odor learning with names (labeling conditions), one condition of odor learning without names (sensory familiarization), and one condition of no learning prior to testing (control conditions). The labeling conditions differed from each other regarding label characteristics. The names were those of odor sources (veridical names), those personally generated by subjects (generated names), or those derived from the chemical names of the odorants (chemical names). Subjects were required to learn 20 fixed associations between odors (targets or distractors) and 20 names during two daily sessions. The learning sessions included two identification tests and ended by a verbal memory test in which subjects recalled odor names. The odor memory test was split into two parts separated by a retention interval of either 20 min (short-term memory) or 24 h (long-term memory). Data showed that olfactory recognition memory was enhanced in subjects who associated veridical or generated names to odors during the learning session. Chemical names were not appropriate to facilitate odor recognition. Similarly, the level of odor identification was higher for veridical and generated names than for chemical names, though the level of verbal memory for chemical names was substantial. Recognition response latencies were systematically longer for a target odor implying a positive response than for a distractor odor implying a negative response. Together, these data suggest that odor recognition and identification are sensitive to the semantic content of labels associated with odors. Odor memory was adversively influenced by time, but this influence was less pronounced when the names were endowed with a rich semantic content.