I had a patient with a chief complaint of deja vu, so I looked it up. (BTW: turned out to be undiagnosed dementia in my patient's case.)

Best article I found was Wild E. J Neurol. 2005 Jan;252(1):1-7. Summarized (and all below references come from Wild's paper):

Definition: "any subjectively inappropriate impression of familiarity of a present experience with an undefined past." (Neppe, 1983)

Wild explains that the definition here is importantly specific: "subjectively inappropriate" means that the patient understands the familiarity is impossible (i.e. this is not a delusion.) "Undefined past" is a non-existent past, and the patient never pinpoints it (because it never happened).

Some theories:

Wigan (1985), Jensen (1868) and Maudsley (1889): a "loss of synchronicity" between two hemispheres of brain, so that they are working "separately but synchronously." Jensen also suggests it is familiarity of one part of the experience generalized to the whole.

Gestalt psychology: object-affect entities. An experience causes an affect, which is identical to the affect assocaited with an unrelated event in the past. Your brain interprets it as a rememberance of the object (as opposed to the affect.)

De Nayer:"tape recorder hypothesis:" you are remembering the event and recollecting the event at the extact same time.

Freud: the situation is similar to a suppressed fantasy, so the fantasy activates as a wish to make improvements in the current situation, so in essence it is the wish to turn back time.

Biology:

Associations with temporal lobe epilepsy suggested that that was the relevant neuroanatomy, but some interesting experiments suggest that deja vu is associated with limbic structures (especially hippocampus and amygdala) and not the temporal neocortex itself. Bancaud (1994) tried to synthesize the available information and proposed that the perception is encoded in the temporal neocortex and remembered in the hippocampus; affective memory is supplied by the amygdala. Then, these relate back to the temporal neocortex as a daja vu. Thus, a situation, as it is experienced and "recorded" (see De Nayer) activates deeper memory structures.

All of this appears to be lateralized to the to the temporal lobe ipsilateral to the dominant hand (which is really the non-dominant hemisphere-- it crosses). In fact, the point of all this was to use the symptom of deja vu to predict where a seizure focus would be. (Unfortuantely, a PET scan study could not confirm this lateralization.)
Because of the prominent association of deja vu with autosomal dominant "partial epilepsy with auditory features" a genetic contribution is suggested. This gene is the LGI1/epitempin gene (10q24).

Assessment should rule out depersonalization or flashbacks. If deja vu is short or infrequent, it is probably normal; but if recurrent, prolonged, or associated with physical sensations, consider TLE. If associated with anxiety or depression, consider psychiatric causes.

Other articles:

A study of 24 epileptics receiving direct stereotactic electrical stimualtion of the brain found that stimualtion of the entorhinal cortex produced more deja vu than the amygdala or hippocampus, while the perirhinal cortex was associated with the recollection of memories.

A prospective study of TLE patients and their "auras" (called simple partial seizures, i.e. seizures with no loss of consciousness, and include deja vu, weird tastes, etc)) found that the SPSs, deja vu, a warm sensation, a cephalic sensation, taste hallucination, and a "strange" sensation predicted an abnormal amygdala ipsliateral to the seizure focus. Though fear was the most common for all TLE patients, the single best predictor of an abnormal amygdala was... deja vu. As described by Wild, this occurred most commonly on the right (i.e. ipsilateral to handedness, or nondominant for language.)
A study of 14 patients with varying frontal lobe damage were tested for a number of memopry parameters; those with incorrect "feeling-of-knowing" had damage in the right prefrontal cortex.

Although deja vu is supposedly distinct from psychosis, and not related to dopamine, there's a case report of a 39yo physician-patient who took amantadine and phenylpropanolamine to ward off the flu, and got intense deja vus. They stopped when he stopped the medications. The authors find other case reports finding the same.

An interesting case report (two cases, actually) found that, contrary to the popular understanding that the deja vu shares some similarity to an actual past event, on formal testing these two patients had recollections which were unlikely related to previous familiarity ("incorrectly recognised low frequency words."). They then created justifications, i.e. they confabulated the recollections.

As an aside, there are 85 articles in Pubmed with the words "deja vu all over again" in the titles. Almost none actually had anything to do with deja vu. That's creativity for you.