Thank you for the reply.

I'll try to create a small input docx file to demonstrate the problem.

Indeed gug.native contains two instances of Header 0, as seen in this snippet:

...
,BlockQuote
 [Para [Str "Appendix"]]
,Para [Str "A",Space,Str "Instructions",Space,Str "for",Space,Str "Adding",Space,Str "an",Space,Str "Appendix",Space,Str "29"]
,Header 0 ("figures",[],[]) [Str "FIGURES"]
,Para [Strong [Str "Figure",Space,Str "Page"]]
,Para [Strong [Str "No",Space,Str "table",Space,Str "of",Space,Str "figures",Space,Str "entries",Space,Str "found."]]
,Header 0 ("tables",[],[]) [Str "TABLES"]
,Para [Strong [Str "Table",Space,Str "Page"]]
,Para [Strong [Str "No",Space,Str "table",Space,Str "of",Space,Str "figures",Space,Str "entries",Space,Str "found."]]
...

In a re-post of the original message, I reported that the problem seems to occur only when writing a reStructuredText file. Converting the same docx file to another format (e.g., odt or html) works without complaint. (I'd tried to delete the original post and leave only the re-post, but it seems that didn't take effect, or didn't take effect in time.)

On Wednesday, August 2, 2017 at 6:55:34 AM UTC-7, John MacFarlane wrote:

That's certainly a bug in pandoc, not a mistake on your part!

I have a guess about what's going on. It would be easiest
to test if you could send or link to an input docx file
sufficient to reproduce the problem. If you can't do that,
could you try the following experiment?

pandoc -f docx -t native -o gug.native gug.docx

Then search gug.native to see if it contains

Header 0

at any point...