Personas are pegs
Picking up on a post from Mr Budd's blog about personas (and leaving the Jason-vs-Andy battle in the comments aside for a second)...
Personas are an interesting technique, and probably one of the most visible initial outputs of a modern UCD process. "Right, first thing we'll do is sort out who are users are". I've seen good design teams (from FP and other folks) deliver quite detailed documentation about key users of a product, and use these to frame conversations or thinking about the ongoing development of said product. They also produce nice-looking customer-friendly documentation early on, which I suspect contributes to their popularity to some degree.
But they ain't magic pixie dust. To my mind, personas are just pegs on which you hang a set of assumptions: "Jeff is 16 years old, likes Linkin Park, lives with his mum and uses a Motorola RAZR on a pay-as-you-go-contract" might be a brief summing-up of a potential end-user, but how does this help us design a service for Jeff? The real detail of his life, the minutiae on which we'll unconsciously make a lot of design decisions, is in our own minds - and in the assumptions we make about Jeff are therefore based on our understanding of him. If we haven't understood what parts of Jeff relate to the product we're designing for him (and this might be very hard indeed when you consider all the psychological, economic, social or other factors which can have an effect), then Jeff runs the risk of becoming a placeholder for our own misconceptions, biases and misunderstandings about him.
There's an analogy I'm seeing here with the use of short stories for expressing requirements in agile-land. Right now at FP we're expressing requirements for the products we're building with short written onto a 6"x4" index card. Working in such limited space requires brevity and certainly doesn't give you room enough to go into detail of a feature - which is the whole point. Devi related to me a comment made by Jeff Patton at XP Day earlier this year, concerning story cards: they're "placeholders for conversations", rather than an expression of every detail. I think the same can be said for personas.
So getting round to my point: personas are placeholders for assumptions about users, and as such cannot be valid unless they're drawn from detailed research and strong understanding of these users. Without this understanding, you implicitly substitute your own biases.