The whole point of steganography is to avoid detection. A "universal test" for steganography would be a constructive proof that steganography is not possible. However, no proof (constructive or not) of the impossibility of steganography is currently known. This implies that no "universal test" for steganography is currently known -- and it is not known whether such a test is even possible.
We can still say a few things. For instance, steganography relies on:
Variability in data. There must be some room for encoding the hidden message. If the complete message format is fixed down to the last bit, then there is no way to include an extra message. That variability must not alter the apparent meaning of the data.
A secret convention. The recipient of the data must know that there is a hidden message, and how to find it.
Among the tools for steganography, an important one is encryption: symmetric encryption has the ability to transform arbitrary data into a sequence of bits which will have the same probability distribution as random noise. The message recipient, of course, knows the decryption key: it is part of the secret convention. Thus, a good steganography tool will begin by an encryption layer. A consequence is that if the medium is in a format which "naturally" allows for some random noise to appear (e.g. a photograph), then steganography will be possible and very hard to detect, even if the method is completely known (because nothing looks more like random noise than random noise).
A common tool against steganography is compression. The basic premise of lossy compression (as is used for media files, e.g. MP3 or JPG) is that irrelevant details can be removed from the file, where "irrelevant" means "does not alter the perceived meaning of the data". Random noise will be tracked and removed by compression. Therefore, compression tends to be at odds with steganography. If you write a filter which automatically recompresses (aggressively) all pictures sent by email, then you will not detect steganography, but you will have made it much harder. In that sense, steganography shares some characteristics with watermarking.
However, some kinds of steganography seem impossible to detect and defeat. For instance, I can use the following convention with a correspondent: next week, I will send him a binary information (a "yes/no") hidden in a photograph of a cat; the binary information will be "yes" if the cat looks to the left of the photograph, and "no" if the cat looks to the right.