Five studies explored identification of odors as an aspect of semantic
memory. All dealt in one way or another with the accessibility of acq
uired olfactory information. The first study examined stability and sh
owed that, consistent with personal reports, people can fail to identi
fy an odor one day yet succeed another. Failure turned more commonly t
o success than vice versa, and once success occurred it tended to recu
r. Confidence ratings implied that subjects generally knew the quality
of their answers. Even incorrect names, though, often carried conside
rable information which sometimes reflected a semantic and sometimes a
perceptual source of errors. The second study showed that profiling o
dors via the American Society of Testing and Materials list of attribu
tes, an exercise in depth of processing, effected no increment in the
identifiability/accessibility beyond an unelaborated second attempt at
retrieval. The third study showed that subjects had only a weak abili
ty to predict the relative recognizability of odors they had failed to
identify. Whereas the strength of the feeling that they would 'know'
an answer if offered choices did not associate significantly with perf
ormance for odors, it did for trivia questions. The fourth study demon
strated an association between ability to discriminate among one set o
f odors and to identify another, but this emerged only after subjects
had received feedback about identity, which essentially changed the ta
sk to one of recognition and effectively stabilized access. The fifth
study illustrated that feedback improves performance dramatically only
for odors involved with it, but that mere retrieval leads to some imp
rovement. The studies suggest a research agenda that could include sup
plemental use of confidence judgements both retrospectively and prospe
ctively in the same subjects to indicate the amount of accessible sema
ntic information; use of second and third guesses to examine subjects'
simultaneously held hypotheses about identity; use of category cuing
or similar techniques to discover the minimum semantic information nee
ded to precipitate identification; some use of subjects trained in qua
ntitative descriptive analysis to explore whether such training enhanc
es semantic memory; and judicious use of mixtures to explore perceptua
l versus semantic errors of identification.