These are set of guidelines suggested by Krzysztof C and Microsoft Design Guidelines on the web on Constant Usage.

1. If a value is constant in nature (never changes its value in its life time) make it a constant, such as Pi. [MSDN2: Microsoft design Guidelines]

2. If a literal used in two different assemblies make it public static read-only.

3. If a literal used in two different types (e.g. classes) in the same assembly make it a public constant and introduce a separate Constants code file.

4. If a literal used in two different members (method or a property) in a same type (e.g. class) make it a private field constant.

5. If a literal occurred in two places in a member, make it a local constant.

6. If a string literal is used at least once, consider it making a constant. (This guideline is the controversial oneJ)

It also says that, using constants is better than not using them.

I also see that the Maintainability aspect should get the precedence over any other “good” technical practices behind constants.