Please see this picture:

http://imgur.com/a/cQW6b

Why are all the genes "crunched" up at 0.9-1.0? I see the line down at 0.4, before I removed one duplicated row, it was down at 0.

Can it be some more trouble with my probes?

The power of 20 is fairly high but not unreasonably so if you are constructing a signed network and you have less than 20 samples). Otherwise I suggest you follow the advice in WGCNA FAQ, item 6.

https://labs.genetics.ucla.edu/horvath/CoexpressionNetwork/Rpackages/WGCNA/faq.html

As I said, there's nothing apriori wrong with your dendrogram, no need to fiddle with the soft thesholding power just to make it look "prettier" - biostatistics is not a beauty contest!

It means you have a pair of genes with high similarity (dissimilarity 0.4, so similarity 0.6) while all other genes have higher dissimilarities. Apriori nothing wrong with that although in most gene expression studies I do see somewhat smaller dissimilarities (many genes pairs reaching down to 0.75 or 0.8). Note that the heights (dissimilarities) also depend strongly on the soft-thresholding power which you may have set a bit too high.

Thank you! I set the soft-thresholding power to 20: http://imgur.com/a/vxkPd

Perhaps I should choose 3 instead: http://imgur.com/a/WRVQx