Science-- there's something for everyone

Thursday, October 10, 2013

Categorizing smells

Our taste receptors can only detect five or six flavors: sweet, salt, bitter, sour, umami and possibly fat. Our eyes have only three types of color detecting cones.  For example, wavelengths of 475 nm trigger our blue cones. Yet smells seemed to defy discrete categorization. Until now.

Jason Castro from Bates College, Arvind Ramanathan from Oak Ridge National Laboratory, and Chakra Chennubhotla from the University of Pittsburg used data from smell tests conducted with dozens of trained olfactory scientists and perfumers. The volunteers rated how well each compound fit one of 146 descriptors, such as 'pleasantness', 'sweetness' or 'heaviness'. The researchers were able to develop a mathematical model of all the kinds of odors we can smell. According to them, we can detect ten broad categories of odor. I know you’re dying to know what they are, so here’s the list:

  1. Fragrant
  2. Woody/resinous
  3. Fruity (non-citrus)
  4. Chemical
  5. Minty/peppermint
  6. Sweet
  7. Popcorn
  8. Lemon
  9. Pungent
  10. Decayed

The last two are both ‘sickening’ odors that you would wish to avoid. 

This shouldn’t be interpreted to mean that everything smells like either lemons or popcorn (though I’m amazed that those two got their own categories). Rather, these are just names the authors picked for broad categories of smells. You can see a list of some of those scent categories below.

Table 1 10 largest-valued descriptors for each of the 10 basis vectors obtained from non-negative matrix factorization.
10 largest-valued descriptors for each of the 10 basis vectors obtained from non-negative matrix factorization.

To be clear, the researchers were classifying single odors, not jambalaya. I wouldn't put too much stock in the names or rigidity of these categories. That said, I find it interesting that there could be discrete divisions of smells. 


Castro JB, Ramanathan A, & Chennubhotla CS (2013). Categorical dimensions of human odor descriptor space revealed by non-negative matrix factorization. PloS one, 8 (9) PMID: 24058466.