UPDATE: as of Scala-2.10, using equals sign is preferred. Old answer:

Methods which return Unit should always use the non-equals syntax. This avoids potential mistakes in implementation carrying over into the API. For example, you could have accidentally done something like this:

object HelloWorld {
  def main(args: Array[String]) = {
    println("Hello!")
    123
  }
}

Trivial example of course, but you can see how this might be a problem. Because the last expression does not return Unit, the method itself will have a return type other than Unit. This is exposed in the public API, and might cause other problems down the road. With the non-equals syntax, it doesn’t matter what the last expression is, Scala fixes the return type as Unit.

It’s also two characters cleaner. 🙂 I also tend to think that the non-equals syntax makes the code just a little easier to read. It is more obvious that the method in question returns Unit rather than some useful value.

On a related note, there is an analogous syntax for abstract methods:

trait Foo {
  def bar(s: String)
}

The method bar has signature String=>Unit. Scala does this when you omit the type annotation on an abstract member. Once again, this is cleaner, and (I think) easier to read.