Via UXMatters, apparently, this is the best way to structure a form for usability:

Form Test

It includes:

  1. NOT using "bolded" labels
  2. Using labels on top of the form fields
  3. Rejecting the common idea of right aligned, left-of-field labels

and these are their conclusions:

Placing a label above an input field works better in most cases, because users aren’t forced to look separately at the label and the input field. Be careful to visually separate the label for the next input field from the previous input field.

and

Reading bold labels is a little bit more difficult for users, so it’s preferable to use plain text labels. However, when using bold labels, you might want to style the input fields not to have heavy borders.

This is one of those studies that really flies in the face of convention, and I'm a guy who tends to think of usability in terms of convention, i.e. the more conventional, the more usable. However, I am impressed with the study - the measured eyetracking with real users on real forms, tested the time between reading the labels and filling out the fields and really did a bang up job all around. Reading the full study and comments is worthwhile (even with that horrible purple background on the page).

What do you think? Do we have a revolution in form usability, or is this something that every website needs to test individually to measure results?