David answered the key point: we don't know exactly, but look at them individually to see what is wrong. If you can legally share the book, upload it or link to it. Otherwise, you can send a PM to me and we can arrange something else.
From what I know, the problem is very unlikely to be CSS. The device shouldn't be reading the CSS at the point it rejects the file. It should be reading the OPF and NCX files and anything else it needs to find those. I am fairly sure the device is only processing that these files are valid. I don't believe it checks if the files referred to in them exist. That is definitely the case for the NCX, but I don't know about the manifest in the OPF.
On the occasions when I see this or when someone sends me a problem book, I start simple. I open it with the calibre viewer to see if it thinks it is an epub at all. This isn't always useful as the calibre viewer is fairly forgiving for errors. Then I either unzip and inspect the contents or open it in Sigil. As David mentioned, Sigil has an error checker to test. If I can't fix the errors in Sigil, or there are to many, or if the code looks horrible, I do an epub-to-epub conversion in calibre. That usually fixes the epubs enough to be put onto the device. But, I usually edit in Sigil and clean up the code.
Alternatives to the above are
EpubCheck to check for errors or
epubfixer to make the changes. I have used both in the past, but prefer my process above.