There has recently been a re-evaluation of our sense of smell, which is now considered a very sensitive and discriminating sense modality by scientists and philosophers. However, the consensus in the literature is that humans, and certainly Western subjects, are very poor at identifying smells: they produce the 'veridical label' for an odour in just 30-50% of cases and there is wide inter-subjective variation in their responses. This suggests that we rarely know what we smell. Is this the right conclusion to draw from the evidence? This paper takes a closer look at the empirical evidence on the smell naming performance of Western subjects and argues that a comparative model of olfactory language and categorisation is more effective at explaining the evidence than a model on which each smell kind is supposed to correspond to one label. One result of applying a comparative model is that we are not quite as poor at naming smells as the commonly cited data would suggest. Another result is a better understanding of the kinds of knowledge we may gain by smelling and how these relate to the linguistic resources, experiences, and practices of different speakers and communities.