The benefits of antidepressants relative to placebos appear to be correlated with the severity of depression (see http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=11799341&dopt=Abstract).
The evidence is complicated by the fact that subjects often know whether they are taking placebos (see http://depression.about.com/library/weekly/aa072202.htm).
Also, it should be noted that existing measures of depression are imperfect. Here's a study that shows an inverse correlation between suicide rates and use of Prozac/SSRIs:http://medicine.plosjournals.org/perlserv/?request=get-document&doi=10.1371/journal.pmed.0030190.
It appears to me that SSRIs (and maybe other antidepressants) were initially used in cases where they were clearly beneficial, and then the use expanded to a much larger set of cases where the evidence is ambiguous.
I have some experience with an SSRI. It took me 2 hours to decide that it was very different from a placebo, but it took me 2 months to decide that I didn't like the net effects. In particular, it caused reduced obsessiveness that was beneficial in a significant fraction of contexts (i.e. when sleeping), but not beneficial as a permanent effect.