If you haven’t already, it’s strongly recommended to read the introduction first, as it introduces the various test types.
There’s also a load of general guidelines that apply to all tests.
There are four main test types:
Reftests should be used to test rendering and layout. They consist of two or more pages with assertions as to whether they render identically or not.
Visual tests should be used for checking rendering where there is a large number of conforming renderings such that reftests are impractical. They consist of a page that renders to final state at which point a screenshot can be taken and compared to an expected rendering for that user agent on that platform.
Manual tests are used as a last resort for anything that can’t be tested using any of the above. They consist of a page that needs manual interaction or verification of the final result.
In general, there is a strong preference towards the first two test types (as they can be easily run without human interaction), so they should be used in preference to the others even if it results in a somewhat cumbersome test; there is a far weaker preference between the first two, and it is at times advisable to use testharness.js tests for things which would typically be tested using reftests but for which it would be overly cumbersome.
In addition to the four main test types, there are also WebDriver tests, which are used exclusively for testing the WebDriver protocol itself. There is currently no documentation about these tests, however.