Good idea to move this, it indeed otherwise doesn't show up in the built docs.

The part in Notes on "There are three main sections" seems to contradict to some extent the first paragraph of the class docstring.

The paragraph on PyNIO and pynetcdf can remain in the module docstringI think, with a single sentence in the class docstring on this topic being enough. An addition to that part is that this module is derived from and almost equal to the standalone package pupynere. The part on Scientific is not relevant anymore probably.

I added some corrections from your remarks.

About the part in Notes on "There are three main sections" , I see the point. The original author uses the same word (attributes) to designate two differents things. At the top of the docstring, attribute means attribute of an (python) object while attribute in the notes refers to the sementic used in netcdf specifications (see http://www.unidata.ucar.edu/software/netcdf/docs/netcdf.html#Dataset-Components).
Both sentences are right IMHO. The module implements the netcdf attributes as illustrated in examples (units is an attribute).
To be accurate, netcdf attribute handling is implemented in netcdf_variable(). This is suggested by the following the NetCDF data structure list. Split the notes into two just for that would be less efficient I think.
Nevertheless, if you have a better idea to clarify that, i'm all ears. :)

I agree with your second point. I left the sentence on scientific.io api in the module docstring.

Thanks for the careful review.

You're right, the attributes usage is slightly confusing but correct. Not sure it's easy to change, so I'll merge this now.

Merged, thanks!