It depends on several things.

1) Why do you build a kernel?

Most of the time, people develop kernels to learn how they work and to 
have fun. In this case just go with any language you know and like.

If you have a favorite language that is not C or C++, for instance a 
functional or a dynamic language, then by any means use it since 1) it 
will make your project interesting to its community and 2) it might 
teach you about some of its limits.

You will probably need to know C though, since you will want to look at 
how Open Source OSs have solved your problems...

On the other hand, if this kernel is not a toy project, you will have 
goal such as limiting the risk of bugs and security issues, working with 
other people... that can help you choose the right language. Obviously 
the most history-tested one is C so if you chose something radically 
different make sure you are not overlooking problems.

2) Target hardware & portability

Does the OS have to be portable? If yes then you need either a language 
that has been ported to a variety of platforms (such as ANSI C or maybe 
Java) or a language that can be ported / re-implemented easily (such as 
Forth). You will want as little Assembly as possible.

On the other hand, if you are targeting a specific, embedded platform, 
maybe using low-level languages (C, Assembly) is the way to go.

3) What kind of kernel are you building?

Micro-kernels such as L4 or Minix3 are not the same kind of beasts as 
the Linux or Windows kernels. The main difference to me is that a single 
programmer can have the whole code base of a micro-kernel in his head, 
whereas it is almost impossible in a monolithic kernel.

What that changes is that in a monolithic kernel you have to limit 
coupling as much as possible, and you need to standardize more. There 
are Assembly language implementations of L4 but I doubt it would be easy 
to implement the Linux kernel in Assembly.

4) What about me?

I do not intend to implement a kernel soon and even if I did I would 
fall into the "for learning and fun" category. I have already 
implemented a bootloader in x86 ASM (also for fun) a few years ago 
though.

If I decided to implement a kernel today it would probably be either in 
C and Lua (because eLuaBrain fascinates me and I wonder if it is 
possible to build most of a kernel in a dynamic language with 
coroutines) or Forth (because it's easy to boostrap and things like 
OpenFirmware fascinate me too).