Visualise a protein as beads on a string (Synteny :) that are alternately sticky or charged. This is the HP model - hydrophobic and polar amino acids only.

This string starts as a fluctuating coil like a sea snake but the repulsion of some charges and the attraction between others combine with the sticky patches to compact the string. Several schools of thought exist on how  this occurs, including hydrophobic collapse and the funnel hypothesis.

Most importantly, there is a difference between small and large proteins. Large ones may fold in several stages, possibly even by domain association. Smaller proteins often go from unfolded to folded (or native state) in one step - and very quickly too.

The time scale of folding is on the order of milliseconds, wheras the famous Levinthal paradox concluded that a random search would take many aeons. Therefore, some sort of ordered pathway must exist - such as hierarchical folding.