How does one find the start of the “Central Directory” in zip files?

Wikipedia has an excellent description of the ZIP file format, but the "central directory" structure is confusing to me. Specifically this:

This ordering allows a ZIP file to be created in one pass, but it is usually decompressed by first reading the central directory at the end.

The problem is that even the trailing header for the central directory is variable length. How then, can someone get the start of the central directory to parse?

(Oh, and I did spend some time looking at APPNOTE.TXT in vain before coming here and asking :P)